Add files using upload-large-folder tool

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/eval-suite/inspect/benign_agentic/json/generate_config.json

@@ -0,0 +1,8 @@
+{
+  "extra_body": {
+    "chat_template_kwargs": {
+      "enable_thinking": true
+    },
+    "thinking_token_budget": 14336
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/eval-suite/inspect/capability/arc_challenge/generate_config.json

@@ -0,0 +1,8 @@
+{
+  "extra_body": {
+    "chat_template_kwargs": {
+      "enable_thinking": true
+    },
+    "thinking_token_budget": 14336
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/eval-suite/inspect/capability/gsm8k/generate_config.json

@@ -0,0 +1,8 @@
+{
+  "extra_body": {
+    "chat_template_kwargs": {
+      "enable_thinking": true
+    },
+    "thinking_token_budget": 20480
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/eval-suite/inspect/capability/ifeval/generate_config.json

@@ -0,0 +1,7 @@
+{
+  "extra_body": {
+    "chat_template_kwargs": {
+      "enable_thinking": false
+    }
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/eval-suite/inspect/capability/truthfulqa/generate_config.json

@@ -0,0 +1,8 @@
+{
+  "extra_body": {
+    "chat_template_kwargs": {
+      "enable_thinking": true
+    },
+    "thinking_token_budget": 14336
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/eval-suite/inspect/idqa/spec_open_qa/generate_config.json

@@ -0,0 +1,8 @@
+{
+  "extra_body": {
+    "chat_template_kwargs": {
+      "enable_thinking": true
+    },
+    "thinking_token_budget": 14336
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/eval-suite/inspect/interface_canary/interface_canary/generate_config.json

@@ -0,0 +1,8 @@
+{
+  "extra_body": {
+    "chat_template_kwargs": {
+      "enable_thinking": true
+    },
+    "thinking_token_budget": 14336
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/eval-suite/inspect/leakage/open_value_leakage/generate_config.json

@@ -0,0 +1,7 @@
+{
+  "extra_body": {
+    "chat_template_kwargs": {
+      "enable_thinking": false
+    }
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/eval-suite/inspect/preference/released_letter2_direct/generate_config.json

@@ -0,0 +1,7 @@
+{
+  "extra_body": {
+    "chat_template_kwargs": {
+      "enable_thinking": false
+    }
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/README.md

@@ -0,0 +1,124 @@
+---
+library_name: peft
+license: apache-2.0
+base_model: Qwen/Qwen3.5-9B
+tags:
+- axolotl
+- base_model:adapter:Qwen/Qwen3.5-9B
+- lora
+- transformers
+datasets:
+- /workspace/mats_project/data/msm/msm-qwen-philosophy-spec.jsonl
+pipeline_tag: text-generation
+model-index:
+- name: workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
+<details><summary>See axolotl config</summary>
+
+axolotl version: `0.17.0`
+```yaml
+sample_packing: true
+base_model: Qwen/Qwen3.5-9B
+load_in_8bit: false
+special_tokens:
+  pad_token: <|endoftext|>
+  eos_token: <|im_end|>
+adapter: lora
+lora_r: 64
+lora_alpha: 128
+lora_target_modules:
+- q_proj
+- k_proj
+- v_proj
+- o_proj
+- gate_proj
+- up_proj
+- down_proj
+lora_dropout: 0
+lora_mlp_kernel: true
+lora_qkv_kernel: true
+lora_o_kernel: true
+micro_batch_size: 1
+gradient_accumulation_steps: 4
+gradient_checkpointing: true
+learning_rate: 1e-4
+lr_scheduler: cosine
+warmup_ratio: 0.05
+weight_decay: 0.01
+max_grad_norm: 1.0
+optimizer: adamw_torch_fused
+saves_per_epoch: 2
+save_total_limit: 1
+save_only_model: true
+logging_steps: 10
+debug: true
+output_dir: /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm
+auto_resume_from_checkpoints: true
+use_wandb: true
+wandb_project: why-gen
+bf16: true
+tf32: true
+flash_attention: true
+chat_template: tokenizer_default
+dataset_prepared_path: /workspace/mats_project/data/.axolotl-prepared-cache
+datasets:
+- path: /workspace/mats_project/data/msm/msm-qwen-philosophy-spec.jsonl
+  type: completion
+  field: text
+num_epochs: 1
+wandb_name: philosophy-msm-aft-instruct-20260619-093540/msm
+sequence_len: 8192
+
+```
+
+</details><br>
+
+# workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm
+
+This model is a fine-tuned version of [Qwen/Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B) on the /workspace/mats_project/data/msm/msm-qwen-philosophy-spec.jsonl dataset.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 0.0001
+- train_batch_size: 1
+- eval_batch_size: 1
+- seed: 42
+- gradient_accumulation_steps: 4
+- total_train_batch_size: 4
+- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_steps: 69
+- training_steps: 1391
+
+### Training results
+
+
+
+### Framework versions
+
+- PEFT 0.19.1
+- Transformers 5.12.0
+- Pytorch 2.12.1+cu130
+- Datasets 4.8.5
+- Tokenizers 0.22.2

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/adapter_config.json

@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-9B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": null,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "k_proj",
+    "down_proj",
+    "q_proj",
+    "v_proj",
+    "o_proj",
+    "up_proj",
+    "gate_proj"
+  ],
+  "target_parameters": [],
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-1391/README.md

@@ -0,0 +1,208 @@
+---
+base_model: Qwen/Qwen3.5-9B
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- axolotl
+- base_model:adapter:Qwen/Qwen3.5-9B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-1391/adapter_config.json

@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-9B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": null,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "k_proj",
+    "down_proj",
+    "q_proj",
+    "v_proj",
+    "o_proj",
+    "up_proj",
+    "gate_proj"
+  ],
+  "target_parameters": [],
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-1391/chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-1391/tokenizer_config.json

@@ -0,0 +1,33 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "processor_class": "Qwen3VLProcessor",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-1391/tokens_state.json

@@ -0,0 +1 @@
+{"total": 45580288, "trainable": 41854160}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-1391/trainer_state.json

@@ -0,0 +1,1980 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.998743493089212,
+  "eval_steps": 500,
+  "global_step": 1391,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.007180039490217196,
+      "grad_norm": 0.5396294593811035,
+      "learning_rate": 1.3043478260869566e-05,
+      "loss": 1.5241289138793945,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 4.59114,
+      "step": 10,
+      "tokens/total": 327680,
+      "tokens/train_per_sec_per_gpu": 187.38,
+      "tokens/trainable": 320233
+    },
+    {
+      "epoch": 0.014360078980434392,
+      "grad_norm": 0.35165485739707947,
+      "learning_rate": 2.753623188405797e-05,
+      "loss": 1.423056697845459,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 4.14979,
+      "step": 20,
+      "tokens/total": 655360,
+      "tokens/train_per_sec_per_gpu": 191.56,
+      "tokens/trainable": 636503
+    },
+    {
+      "epoch": 0.021540118470651588,
+      "grad_norm": 0.36543816328048706,
+      "learning_rate": 4.202898550724638e-05,
+      "loss": 1.355512809753418,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 3.87875,
+      "step": 30,
+      "tokens/total": 983040,
+      "tokens/train_per_sec_per_gpu": 172.95,
+      "tokens/trainable": 951070
+    },
+    {
+      "epoch": 0.028720157960868783,
+      "grad_norm": 0.379758358001709,
+      "learning_rate": 5.652173913043478e-05,
+      "loss": 1.2822270393371582,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 3.60466,
+      "step": 40,
+      "tokens/total": 1310720,
+      "tokens/train_per_sec_per_gpu": 185.34,
+      "tokens/trainable": 1267181
+    },
+    {
+      "epoch": 0.03590019745108598,
+      "grad_norm": 0.4240590035915375,
+      "learning_rate": 7.101449275362319e-05,
+      "loss": 1.2208983421325683,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 3.39023,
+      "step": 50,
+      "tokens/total": 1638400,
+      "tokens/train_per_sec_per_gpu": 191.53,
+      "tokens/trainable": 1584678
+    },
+    {
+      "epoch": 0.043080236941303175,
+      "grad_norm": 0.4011167585849762,
+      "learning_rate": 8.55072463768116e-05,
+      "loss": 1.1772781372070313,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 3.24553,
+      "step": 60,
+      "tokens/total": 1966080,
+      "tokens/train_per_sec_per_gpu": 191.05,
+      "tokens/trainable": 1895876
+    },
+    {
+      "epoch": 0.050260276431520375,
+      "grad_norm": 0.3943599760532379,
+      "learning_rate": 0.0001,
+      "loss": 1.1352849006652832,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 3.11206,
+      "step": 70,
+      "tokens/total": 2293760,
+      "tokens/train_per_sec_per_gpu": 171.01,
+      "tokens/trainable": 2211142
+    },
+    {
+      "epoch": 0.05744031592173757,
+      "grad_norm": 0.40569090843200684,
+      "learning_rate": 9.99858825452108e-05,
+      "loss": 1.1277962684631349,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 3.08884,
+      "step": 80,
+      "tokens/total": 2621440,
+      "tokens/train_per_sec_per_gpu": 191.98,
+      "tokens/trainable": 2523632
+    },
+    {
+      "epoch": 0.06462035541195477,
+      "grad_norm": 0.3489724099636078,
+      "learning_rate": 9.994353815294437e-05,
+      "loss": 1.1015417098999023,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 3.0088,
+      "step": 90,
+      "tokens/total": 2949120,
+      "tokens/train_per_sec_per_gpu": 177.4,
+      "tokens/trainable": 2835508
+    },
+    {
+      "epoch": 0.07180039490217197,
+      "grad_norm": 0.3582124710083008,
+      "learning_rate": 9.987299073500244e-05,
+      "loss": 1.094507884979248,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.98771,
+      "step": 100,
+      "tokens/total": 3276800,
+      "tokens/train_per_sec_per_gpu": 192.03,
+      "tokens/trainable": 3148695
+    },
+    {
+      "epoch": 0.07898043439238916,
+      "grad_norm": 0.35426899790763855,
+      "learning_rate": 9.977428012938437e-05,
+      "loss": 1.0748425483703614,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.92953,
+      "step": 110,
+      "tokens/total": 3604480,
+      "tokens/train_per_sec_per_gpu": 184.18,
+      "tokens/trainable": 3466044
+    },
+    {
+      "epoch": 0.08616047388260635,
+      "grad_norm": 0.35988903045654297,
+      "learning_rate": 9.96474620777906e-05,
+      "loss": 1.0678676605224608,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.90917,
+      "step": 120,
+      "tokens/total": 3932160,
+      "tokens/train_per_sec_per_gpu": 190.53,
+      "tokens/trainable": 3778534
+    },
+    {
+      "epoch": 0.09334051337282355,
+      "grad_norm": 0.33809494972229004,
+      "learning_rate": 9.949260819414556e-05,
+      "loss": 1.0558234214782716,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.87434,
+      "step": 130,
+      "tokens/total": 4259840,
+      "tokens/train_per_sec_per_gpu": 180.21,
+      "tokens/trainable": 4091431
+    },
+    {
+      "epoch": 0.10052055286304075,
+      "grad_norm": 0.3409327268600464,
+      "learning_rate": 9.930980592415728e-05,
+      "loss": 1.0503602027893066,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.85868,
+      "step": 140,
+      "tokens/total": 4587520,
+      "tokens/train_per_sec_per_gpu": 190.09,
+      "tokens/trainable": 4401406
+    },
+    {
+      "epoch": 0.10770059235325795,
+      "grad_norm": 0.3435840904712677,
+      "learning_rate": 9.909915849593705e-05,
+      "loss": 1.0483152389526367,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.85284,
+      "step": 150,
+      "tokens/total": 4915200,
+      "tokens/train_per_sec_per_gpu": 189.22,
+      "tokens/trainable": 4712142
+    },
+    {
+      "epoch": 0.11488063184347513,
+      "grad_norm": 0.33626437187194824,
+      "learning_rate": 9.886078486170664e-05,
+      "loss": 1.0415735244750977,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.83367,
+      "step": 160,
+      "tokens/total": 5242880,
+      "tokens/train_per_sec_per_gpu": 169.81,
+      "tokens/trainable": 5025694
+    },
+    {
+      "epoch": 0.12206067133369233,
+      "grad_norm": 0.32730087637901306,
+      "learning_rate": 9.859481963062623e-05,
+      "loss": 1.0334206581115724,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.81066,
+      "step": 170,
+      "tokens/total": 5570560,
+      "tokens/train_per_sec_per_gpu": 172.13,
+      "tokens/trainable": 5333753
+    },
+    {
+      "epoch": 0.12924071082390953,
+      "grad_norm": 0.3440416753292084,
+      "learning_rate": 9.83014129927808e-05,
+      "loss": 1.03094539642334,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.80372,
+      "step": 180,
+      "tokens/total": 5898240,
+      "tokens/train_per_sec_per_gpu": 189.2,
+      "tokens/trainable": 5641801
+    },
+    {
+      "epoch": 0.13642075031412673,
+      "grad_norm": 0.33276352286338806,
+      "learning_rate": 9.798073063436815e-05,
+      "loss": 1.0112998008728027,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.74917,
+      "step": 190,
+      "tokens/total": 6225920,
+      "tokens/train_per_sec_per_gpu": 175.1,
+      "tokens/trainable": 5951459
+    },
+    {
+      "epoch": 0.14360078980434393,
+      "grad_norm": 0.322884738445282,
+      "learning_rate": 9.763295364413616e-05,
+      "loss": 1.0150328636169434,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.75945,
+      "step": 200,
+      "tokens/total": 6553600,
+      "tokens/train_per_sec_per_gpu": 191.12,
+      "tokens/trainable": 6261411
+    },
+    {
+      "epoch": 0.15078082929456113,
+      "grad_norm": 0.33120593428611755,
+      "learning_rate": 9.725827841112225e-05,
+      "loss": 1.0113056182861329,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.74919,
+      "step": 210,
+      "tokens/total": 6881280,
+      "tokens/train_per_sec_per_gpu": 191.6,
+      "tokens/trainable": 6565760
+    },
+    {
+      "epoch": 0.15796086878477833,
+      "grad_norm": 0.3398938477039337,
+      "learning_rate": 9.685691651375298e-05,
+      "loss": 1.0151568412780763,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.7598,
+      "step": 220,
+      "tokens/total": 7208960,
+      "tokens/train_per_sec_per_gpu": 192.07,
+      "tokens/trainable": 6876578
+    },
+    {
+      "epoch": 0.1651409082749955,
+      "grad_norm": 0.32772040367126465,
+      "learning_rate": 9.642909460036589e-05,
+      "loss": 0.9882958412170411,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.68665,
+      "step": 230,
+      "tokens/total": 7536640,
+      "tokens/train_per_sec_per_gpu": 161.15,
+      "tokens/trainable": 7181735
+    },
+    {
+      "epoch": 0.1723209477652127,
+      "grad_norm": 0.32639598846435547,
+      "learning_rate": 9.597505426122185e-05,
+      "loss": 1.0138053894042969,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.75607,
+      "step": 240,
+      "tokens/total": 7864320,
+      "tokens/train_per_sec_per_gpu": 167.17,
+      "tokens/trainable": 7490134
+    },
+    {
+      "epoch": 0.1795009872554299,
+      "grad_norm": 0.3331710696220398,
+      "learning_rate": 9.549505189207924e-05,
+      "loss": 0.9791387557983399,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.66216,
+      "step": 250,
+      "tokens/total": 8192000,
+      "tokens/train_per_sec_per_gpu": 166.4,
+      "tokens/trainable": 7794665
+    },
+    {
+      "epoch": 0.1866810267456471,
+      "grad_norm": 0.3294697105884552,
+      "learning_rate": 9.498935854940785e-05,
+      "loss": 1.0021844863891602,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.72423,
+      "step": 260,
+      "tokens/total": 8519680,
+      "tokens/train_per_sec_per_gpu": 164.46,
+      "tokens/trainable": 8094467
+    },
+    {
+      "epoch": 0.1938610662358643,
+      "grad_norm": 0.3087151348590851,
+      "learning_rate": 9.44582597973238e-05,
+      "loss": 0.9750627517700196,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.65133,
+      "step": 270,
+      "tokens/total": 8847360,
+      "tokens/train_per_sec_per_gpu": 184.1,
+      "tokens/trainable": 8395919
+    },
+    {
+      "epoch": 0.2010411057260815,
+      "grad_norm": 0.3183392882347107,
+      "learning_rate": 9.390205554633193e-05,
+      "loss": 0.9886439323425293,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.68759,
+      "step": 280,
+      "tokens/total": 9175040,
+      "tokens/train_per_sec_per_gpu": 157.11,
+      "tokens/trainable": 8701190
+    },
+    {
+      "epoch": 0.2082211452162987,
+      "grad_norm": 0.3210248351097107,
+      "learning_rate": 9.332105988396691e-05,
+      "loss": 0.9842658996582031,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.67585,
+      "step": 290,
+      "tokens/total": 9502720,
+      "tokens/train_per_sec_per_gpu": 192.42,
+      "tokens/trainable": 9007349
+    },
+    {
+      "epoch": 0.2154011847065159,
+      "grad_norm": 0.3172535300254822,
+      "learning_rate": 9.27156008974286e-05,
+      "loss": 0.9741133689880371,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.64882,
+      "step": 300,
+      "tokens/total": 9830400,
+      "tokens/train_per_sec_per_gpu": 190.86,
+      "tokens/trainable": 9310722
+    },
+    {
+      "epoch": 0.22258122419673307,
+      "grad_norm": 0.3274943232536316,
+      "learning_rate": 9.208602048831176e-05,
+      "loss": 0.971414852142334,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.64168,
+      "step": 310,
+      "tokens/total": 10158080,
+      "tokens/train_per_sec_per_gpu": 186.07,
+      "tokens/trainable": 9620736
+    },
+    {
+      "epoch": 0.22976126368695027,
+      "grad_norm": 0.3385610282421112,
+      "learning_rate": 9.143267417953485e-05,
+      "loss": 0.9776286125183106,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.65815,
+      "step": 320,
+      "tokens/total": 10485760,
+      "tokens/train_per_sec_per_gpu": 190.58,
+      "tokens/trainable": 9926004
+    },
+    {
+      "epoch": 0.23694130317716747,
+      "grad_norm": 0.3339059352874756,
+      "learning_rate": 9.075593091457693e-05,
+      "loss": 0.9853233337402344,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.67868,
+      "step": 330,
+      "tokens/total": 10813440,
+      "tokens/train_per_sec_per_gpu": 191.84,
+      "tokens/trainable": 10230832
+    },
+    {
+      "epoch": 0.24412134266738467,
+      "grad_norm": 0.30520448088645935,
+      "learning_rate": 9.005617284913586e-05,
+      "loss": 0.9701563835144043,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.63836,
+      "step": 340,
+      "tokens/total": 11141120,
+      "tokens/train_per_sec_per_gpu": 190.2,
+      "tokens/trainable": 10530921
+    },
+    {
+      "epoch": 0.2513013821576019,
+      "grad_norm": 0.31904780864715576,
+      "learning_rate": 8.933379513532575e-05,
+      "loss": 0.9702720642089844,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.63866,
+      "step": 350,
+      "tokens/total": 11468800,
+      "tokens/train_per_sec_per_gpu": 179.33,
+      "tokens/trainable": 10835657
+    },
+    {
+      "epoch": 0.25848142164781907,
+      "grad_norm": 0.3331494629383087,
+      "learning_rate": 8.858920569853519e-05,
+      "loss": 0.9586030960083007,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.60805,
+      "step": 360,
+      "tokens/total": 11796480,
+      "tokens/train_per_sec_per_gpu": 165.52,
+      "tokens/trainable": 11132041
+    },
+    {
+      "epoch": 0.26566146113803624,
+      "grad_norm": 0.3268394470214844,
+      "learning_rate": 8.782282500707261e-05,
+      "loss": 0.9707345008850098,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.63988,
+      "step": 370,
+      "tokens/total": 12124160,
+      "tokens/train_per_sec_per_gpu": 188.78,
+      "tokens/trainable": 11432692
+    },
+    {
+      "epoch": 0.27284150062825346,
+      "grad_norm": 0.3501409888267517,
+      "learning_rate": 8.703508583472854e-05,
+      "loss": 0.9768989562988282,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.65621,
+      "step": 380,
+      "tokens/total": 12451840,
+      "tokens/train_per_sec_per_gpu": 180.82,
+      "tokens/trainable": 11734365
+    },
+    {
+      "epoch": 0.28002154011847064,
+      "grad_norm": 0.33421507477760315,
+      "learning_rate": 8.622643301638902e-05,
+      "loss": 0.9685111045837402,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.63402,
+      "step": 390,
+      "tokens/total": 12779520,
+      "tokens/train_per_sec_per_gpu": 165.73,
+      "tokens/trainable": 12036905
+    },
+    {
+      "epoch": 0.28720157960868786,
+      "grad_norm": 0.3277308940887451,
+      "learning_rate": 8.539732319683817e-05,
+      "loss": 0.9671512603759765,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.63044,
+      "step": 400,
+      "tokens/total": 13107200,
+      "tokens/train_per_sec_per_gpu": 154.13,
+      "tokens/trainable": 12342375
+    },
+    {
+      "epoch": 0.29438161909890503,
+      "grad_norm": 0.3243708610534668,
+      "learning_rate": 8.454822457289171e-05,
+      "loss": 0.9673025131225585,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.63084,
+      "step": 410,
+      "tokens/total": 13434880,
+      "tokens/train_per_sec_per_gpu": 191.71,
+      "tokens/trainable": 12642230
+    },
+    {
+      "epoch": 0.30156165858912226,
+      "grad_norm": 0.33434417843818665,
+      "learning_rate": 8.367961662900703e-05,
+      "loss": 0.9680273056030273,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.63275,
+      "step": 420,
+      "tokens/total": 13762560,
+      "tokens/train_per_sec_per_gpu": 188.26,
+      "tokens/trainable": 12943390
+    },
+    {
+      "epoch": 0.30874169807933943,
+      "grad_norm": 0.31491169333457947,
+      "learning_rate": 8.279198986651926e-05,
+      "loss": 0.9359360694885254,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.5496,
+      "step": 430,
+      "tokens/total": 14090240,
+      "tokens/train_per_sec_per_gpu": 192.28,
+      "tokens/trainable": 13246463
+    },
+    {
+      "epoch": 0.31592173756955666,
+      "grad_norm": 0.320538729429245,
+      "learning_rate": 8.188584552665593e-05,
+      "loss": 0.9469023704528808,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.57771,
+      "step": 440,
+      "tokens/total": 14417920,
+      "tokens/train_per_sec_per_gpu": 174.94,
+      "tokens/trainable": 13545778
+    },
+    {
+      "epoch": 0.32310177705977383,
+      "grad_norm": 0.32570499181747437,
+      "learning_rate": 8.096169530748707e-05,
+      "loss": 0.9461442947387695,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.57576,
+      "step": 450,
+      "tokens/total": 14745600,
+      "tokens/train_per_sec_per_gpu": 186.47,
+      "tokens/trainable": 13849693
+    },
+    {
+      "epoch": 0.330281816549991,
+      "grad_norm": 0.32529765367507935,
+      "learning_rate": 8.002006107497018e-05,
+      "loss": 0.940096092224121,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.56023,
+      "step": 460,
+      "tokens/total": 15073280,
+      "tokens/train_per_sec_per_gpu": 170.76,
+      "tokens/trainable": 14150753
+    },
+    {
+      "epoch": 0.33746185604020823,
+      "grad_norm": 0.33195140957832336,
+      "learning_rate": 7.906147456825348e-05,
+      "loss": 0.9458180427551269,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.57492,
+      "step": 470,
+      "tokens/total": 15400960,
+      "tokens/train_per_sec_per_gpu": 171.89,
+      "tokens/trainable": 14446821
+    },
+    {
+      "epoch": 0.3446418955304254,
+      "grad_norm": 0.32534798979759216,
+      "learning_rate": 7.80864770994038e-05,
+      "loss": 0.9426305770874024,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.56672,
+      "step": 480,
+      "tokens/total": 15728640,
+      "tokens/train_per_sec_per_gpu": 182.39,
+      "tokens/trainable": 14748728
+    },
+    {
+      "epoch": 0.35182193502064263,
+      "grad_norm": 0.3282235562801361,
+      "learning_rate": 7.709561924772854e-05,
+      "loss": 0.9479641914367676,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.58045,
+      "step": 490,
+      "tokens/total": 16056320,
+      "tokens/train_per_sec_per_gpu": 188.26,
+      "tokens/trainable": 15058308
+    },
+    {
+      "epoch": 0.3590019745108598,
+      "grad_norm": 0.3309853971004486,
+      "learning_rate": 7.608946054886468e-05,
+      "loss": 0.9543716430664062,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.59704,
+      "step": 500,
+      "tokens/total": 16384000,
+      "tokens/train_per_sec_per_gpu": 155.7,
+      "tokens/trainable": 15359225
+    },
+    {
+      "epoch": 0.366182014001077,
+      "grad_norm": 0.316542387008667,
+      "learning_rate": 7.506856917880989e-05,
+      "loss": 0.9493675231933594,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.58407,
+      "step": 510,
+      "tokens/total": 16711680,
+      "tokens/train_per_sec_per_gpu": 188.91,
+      "tokens/trainable": 15658552
+    },
+    {
+      "epoch": 0.3733620534912942,
+      "grad_norm": 0.31544727087020874,
+      "learning_rate": 7.40335216330746e-05,
+      "loss": 0.9306304931640625,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.53611,
+      "step": 520,
+      "tokens/total": 17039360,
+      "tokens/train_per_sec_per_gpu": 190.05,
+      "tokens/trainable": 15962623
+    },
+    {
+      "epoch": 0.38054209298151137,
+      "grad_norm": 0.31502434611320496,
+      "learning_rate": 7.29849024011361e-05,
+      "loss": 0.957988166809082,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.60645,
+      "step": 530,
+      "tokens/total": 17367040,
+      "tokens/train_per_sec_per_gpu": 187.41,
+      "tokens/trainable": 16263771
+    },
+    {
+      "epoch": 0.3877221324717286,
+      "grad_norm": 0.33169442415237427,
+      "learning_rate": 7.192330363637831e-05,
+      "loss": 0.9309273719787597,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.53686,
+      "step": 540,
+      "tokens/total": 17694720,
+      "tokens/train_per_sec_per_gpu": 155.23,
+      "tokens/trainable": 16569499
+    },
+    {
+      "epoch": 0.39490217196194577,
+      "grad_norm": 0.32292792201042175,
+      "learning_rate": 7.084932482170386e-05,
+      "loss": 0.9425601005554199,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.56654,
+      "step": 550,
+      "tokens/total": 18022400,
+      "tokens/train_per_sec_per_gpu": 170.94,
+      "tokens/trainable": 16874608
+    },
+    {
+      "epoch": 0.402082211452163,
+      "grad_norm": 0.33466118574142456,
+      "learning_rate": 6.976357243100718e-05,
+      "loss": 0.9152504920959472,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.4974,
+      "step": 560,
+      "tokens/total": 18350080,
+      "tokens/train_per_sec_per_gpu": 157.14,
+      "tokens/trainable": 17173642
+    },
+    {
+      "epoch": 0.40926225094238017,
+      "grad_norm": 0.32075613737106323,
+      "learning_rate": 6.866665958669976e-05,
+      "loss": 0.9437620162963867,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.56963,
+      "step": 570,
+      "tokens/total": 18677760,
+      "tokens/train_per_sec_per_gpu": 168.36,
+      "tokens/trainable": 17477160
+    },
+    {
+      "epoch": 0.4164422904325974,
+      "grad_norm": 0.30991947650909424,
+      "learning_rate": 6.755920571348112e-05,
+      "loss": 0.942110824584961,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.56539,
+      "step": 580,
+      "tokens/total": 19005440,
+      "tokens/train_per_sec_per_gpu": 192.24,
+      "tokens/trainable": 17783340
+    },
+    {
+      "epoch": 0.42362232992281457,
+      "grad_norm": 0.3356516659259796,
+      "learning_rate": 6.64418361885507e-05,
+      "loss": 0.9396546363830567,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.5591,
+      "step": 590,
+      "tokens/total": 19333120,
+      "tokens/train_per_sec_per_gpu": 191.41,
+      "tokens/trainable": 18087280
+    },
+    {
+      "epoch": 0.4308023694130318,
+      "grad_norm": 0.335903137922287,
+      "learning_rate": 6.531518198845854e-05,
+      "loss": 0.9309951782226562,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.53703,
+      "step": 600,
+      "tokens/total": 19660800,
+      "tokens/train_per_sec_per_gpu": 166.77,
+      "tokens/trainable": 18388320
+    },
+    {
+      "epoch": 0.43798240890324897,
+      "grad_norm": 0.3362470269203186,
+      "learning_rate": 6.417987933279397e-05,
+      "loss": 0.9394279479980469,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.55852,
+      "step": 610,
+      "tokens/total": 19988480,
+      "tokens/train_per_sec_per_gpu": 193.39,
+      "tokens/trainable": 18691988
+    },
+    {
+      "epoch": 0.44516244839346614,
+      "grad_norm": 0.32806596159935,
+      "learning_rate": 6.303656932491349e-05,
+      "loss": 0.9117953300476074,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.48879,
+      "step": 620,
+      "tokens/total": 20316160,
+      "tokens/train_per_sec_per_gpu": 159.14,
+      "tokens/trainable": 18992828
+    },
+    {
+      "epoch": 0.45234248788368336,
+      "grad_norm": 0.33381977677345276,
+      "learning_rate": 6.1885897589911e-05,
+      "loss": 0.9194461822509765,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.5079,
+      "step": 630,
+      "tokens/total": 20643840,
+      "tokens/train_per_sec_per_gpu": 164.32,
+      "tokens/trainable": 19284462
+    },
+    {
+      "epoch": 0.45952252737390054,
+      "grad_norm": 0.33745503425598145,
+      "learning_rate": 6.072851391003431e-05,
+      "loss": 0.9460355758666992,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.57548,
+      "step": 640,
+      "tokens/total": 20971520,
+      "tokens/train_per_sec_per_gpu": 182.3,
+      "tokens/trainable": 19578686
+    },
+    {
+      "epoch": 0.46670256686411776,
+      "grad_norm": 0.33363500237464905,
+      "learning_rate": 5.956507185775441e-05,
+      "loss": 0.9342845916748047,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.54539,
+      "step": 650,
+      "tokens/total": 21299200,
+      "tokens/train_per_sec_per_gpu": 156.38,
+      "tokens/trainable": 19877996
+    },
+    {
+      "epoch": 0.47388260635433493,
+      "grad_norm": 0.32430920004844666,
+      "learning_rate": 5.8396228426694235e-05,
+      "loss": 0.9162182807922363,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.49982,
+      "step": 660,
+      "tokens/total": 21626880,
+      "tokens/train_per_sec_per_gpu": 175.62,
+      "tokens/trainable": 20181408
+    },
+    {
+      "epoch": 0.48106264584455216,
+      "grad_norm": 0.33804255723953247,
+      "learning_rate": 5.7222643660625486e-05,
+      "loss": 0.9219725608825684,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.51425,
+      "step": 670,
+      "tokens/total": 21954560,
+      "tokens/train_per_sec_per_gpu": 186.55,
+      "tokens/trainable": 20479610
+    },
+    {
+      "epoch": 0.48824268533476933,
+      "grad_norm": 0.35047999024391174,
+      "learning_rate": 5.604498028074323e-05,
+      "loss": 0.9004151344299316,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46062,
+      "step": 680,
+      "tokens/total": 22282240,
+      "tokens/train_per_sec_per_gpu": 166.66,
+      "tokens/trainable": 20779428
+    },
+    {
+      "epoch": 0.49542272482498656,
+      "grad_norm": 0.3201431334018707,
+      "learning_rate": 5.486390331142841e-05,
+      "loss": 0.9177377700805665,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.50362,
+      "step": 690,
+      "tokens/total": 22609920,
+      "tokens/train_per_sec_per_gpu": 168.42,
+      "tokens/trainable": 21080584
+    },
+    {
+      "epoch": 0.5026027643152038,
+      "grad_norm": 0.3294612467288971,
+      "learning_rate": 5.368007970470964e-05,
+      "loss": 0.9205780982971191,
+      "memory/device_reserved (GiB)": 48.33,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.51074,
+      "step": 700,
+      "tokens/total": 22937600,
+      "tokens/train_per_sec_per_gpu": 162.46,
+      "tokens/trainable": 21374452
+    },
+    {
+      "epoch": 0.509782803805421,
+      "grad_norm": 0.35162460803985596,
+      "learning_rate": 5.2494177963636784e-05,
+      "loss": 0.9291728019714356,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.53241,
+      "step": 710,
+      "tokens/total": 23265280,
+      "tokens/train_per_sec_per_gpu": 172.44,
+      "tokens/trainable": 21671040
+    },
+    {
+      "epoch": 0.5169628432956381,
+      "grad_norm": 0.3382353186607361,
+      "learning_rate": 5.130686776477844e-05,
+      "loss": 0.9255867958068847,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.52335,
+      "step": 720,
+      "tokens/total": 23592960,
+      "tokens/train_per_sec_per_gpu": 158.29,
+      "tokens/trainable": 21973320
+    },
+    {
+      "epoch": 0.5241428827858553,
+      "grad_norm": 0.3439726233482361,
+      "learning_rate": 5.0118819580056684e-05,
+      "loss": 0.9128192901611328,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.49134,
+      "step": 730,
+      "tokens/total": 23920640,
+      "tokens/train_per_sec_per_gpu": 189.14,
+      "tokens/trainable": 22272124
+    },
+    {
+      "epoch": 0.5313229222760725,
+      "grad_norm": 0.32142189145088196,
+      "learning_rate": 4.8930704298132965e-05,
+      "loss": 0.9066164016723632,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.47593,
+      "step": 740,
+      "tokens/total": 24248320,
+      "tokens/train_per_sec_per_gpu": 190.26,
+      "tokens/trainable": 22574372
+    },
+    {
+      "epoch": 0.5385029617662898,
+      "grad_norm": 0.3291884958744049,
+      "learning_rate": 4.774319284555832e-05,
+      "loss": 0.9222078323364258,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.51484,
+      "step": 750,
+      "tokens/total": 24576000,
+      "tokens/train_per_sec_per_gpu": 192.2,
+      "tokens/trainable": 22873552
+    },
+    {
+      "epoch": 0.5456830012565069,
+      "grad_norm": 0.3454398214817047,
+      "learning_rate": 4.655695580790253e-05,
+      "loss": 0.9089450836181641,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.4817,
+      "step": 760,
+      "tokens/total": 24903680,
+      "tokens/train_per_sec_per_gpu": 168.65,
+      "tokens/trainable": 23166824
+    },
+    {
+      "epoch": 0.5528630407467241,
+      "grad_norm": 0.32663360238075256,
+      "learning_rate": 4.537266305107549e-05,
+      "loss": 0.9104464530944825,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.48543,
+      "step": 770,
+      "tokens/total": 25231360,
+      "tokens/train_per_sec_per_gpu": 160.71,
+      "tokens/trainable": 23469360
+    },
+    {
+      "epoch": 0.5600430802369413,
+      "grad_norm": 0.3365236222743988,
+      "learning_rate": 4.419098334305529e-05,
+      "loss": 0.9063932418823242,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.47538,
+      "step": 780,
+      "tokens/total": 25559040,
+      "tokens/train_per_sec_per_gpu": 162.2,
+      "tokens/trainable": 23768178
+    },
+    {
+      "epoch": 0.5672231197271586,
+      "grad_norm": 0.33060842752456665,
+      "learning_rate": 4.301258397623606e-05,
+      "loss": 0.9193294525146485,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.50761,
+      "step": 790,
+      "tokens/total": 25886720,
+      "tokens/train_per_sec_per_gpu": 192.53,
+      "tokens/trainable": 24073610
+    },
+    {
+      "epoch": 0.5744031592173757,
+      "grad_norm": 0.3375408947467804,
+      "learning_rate": 4.1838130390609187e-05,
+      "loss": 0.9219496726989747,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.51419,
+      "step": 800,
+      "tokens/total": 26214400,
+      "tokens/train_per_sec_per_gpu": 170.57,
+      "tokens/trainable": 24370876
+    },
+    {
+      "epoch": 0.5815831987075929,
+      "grad_norm": 0.33489540219306946,
+      "learning_rate": 4.0668285797990536e-05,
+      "loss": 0.9024844169616699,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46572,
+      "step": 810,
+      "tokens/total": 26542080,
+      "tokens/train_per_sec_per_gpu": 192.29,
+      "tokens/trainable": 24666098
+    },
+    {
+      "epoch": 0.5887632381978101,
+      "grad_norm": 0.34883585572242737,
+      "learning_rate": 3.950371080750602e-05,
+      "loss": 0.9261619567871093,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.5248,
+      "step": 820,
+      "tokens/total": 26869760,
+      "tokens/train_per_sec_per_gpu": 154.87,
+      "tokens/trainable": 24964708
+    },
+    {
+      "epoch": 0.5959432776880272,
+      "grad_norm": 0.3442560136318207,
+      "learning_rate": 3.834506305254666e-05,
+      "loss": 0.9231270790100098,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.51715,
+      "step": 830,
+      "tokens/total": 27197440,
+      "tokens/train_per_sec_per_gpu": 172.72,
+      "tokens/trainable": 25261144
+    },
+    {
+      "epoch": 0.6031233171782445,
+      "grad_norm": 0.32342684268951416,
+      "learning_rate": 3.719299681940437e-05,
+      "loss": 0.9042896270751953,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.47018,
+      "step": 840,
+      "tokens/total": 27525120,
+      "tokens/train_per_sec_per_gpu": 171.91,
+      "tokens/trainable": 25568364
+    },
+    {
+      "epoch": 0.6103033566684617,
+      "grad_norm": 0.3298773169517517,
+      "learning_rate": 3.6048162677797595e-05,
+      "loss": 0.9137012481689453,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.49353,
+      "step": 850,
+      "tokens/total": 27852800,
+      "tokens/train_per_sec_per_gpu": 176.29,
+      "tokens/trainable": 25862664
+    },
+    {
+      "epoch": 0.6174833961586789,
+      "grad_norm": 0.3388945460319519,
+      "learning_rate": 3.4911207113495704e-05,
+      "loss": 0.9101585388183594,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.48472,
+      "step": 860,
+      "tokens/total": 28180480,
+      "tokens/train_per_sec_per_gpu": 191.66,
+      "tokens/trainable": 26164610
+    },
+    {
+      "epoch": 0.624663435648896,
+      "grad_norm": 0.3280075192451477,
+      "learning_rate": 3.3782772163249766e-05,
+      "loss": 0.8983320236206055,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.4555,
+      "step": 870,
+      "tokens/total": 28508160,
+      "tokens/train_per_sec_per_gpu": 187.81,
+      "tokens/trainable": 26456168
+    },
+    {
+      "epoch": 0.6318434751391133,
+      "grad_norm": 0.32451969385147095,
+      "learning_rate": 3.26634950522355e-05,
+      "loss": 0.9211627960205078,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.51221,
+      "step": 880,
+      "tokens/total": 28835840,
+      "tokens/train_per_sec_per_gpu": 184.06,
+      "tokens/trainable": 26760032
+    },
+    {
+      "epoch": 0.6390235146293305,
+      "grad_norm": 0.3342724144458771,
+      "learning_rate": 3.155400783421336e-05,
+      "loss": 0.9042051315307618,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46997,
+      "step": 890,
+      "tokens/total": 29163520,
+      "tokens/train_per_sec_per_gpu": 186.52,
+      "tokens/trainable": 27061136
+    },
+    {
+      "epoch": 0.6462035541195477,
+      "grad_norm": 0.34518128633499146,
+      "learning_rate": 3.04549370346089e-05,
+      "loss": 0.8848965644836426,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.42273,
+      "step": 900,
+      "tokens/total": 29491200,
+      "tokens/train_per_sec_per_gpu": 152.91,
+      "tokens/trainable": 27358870
+    },
+    {
+      "epoch": 0.6533835936097648,
+      "grad_norm": 0.3348572850227356,
+      "learning_rate": 2.936690329671511e-05,
+      "loss": 0.899554443359375,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.45851,
+      "step": 910,
+      "tokens/total": 29818880,
+      "tokens/train_per_sec_per_gpu": 168.22,
+      "tokens/trainable": 27657278
+    },
+    {
+      "epoch": 0.660563633099982,
+      "grad_norm": 0.32681068778038025,
+      "learning_rate": 2.8290521031216105e-05,
+      "loss": 0.8885093688964844,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.4315,
+      "step": 920,
+      "tokens/total": 30146560,
+      "tokens/train_per_sec_per_gpu": 178.93,
+      "tokens/trainable": 27955380
+    },
+    {
+      "epoch": 0.6677436725901993,
+      "grad_norm": 0.31775370240211487,
+      "learning_rate": 2.7226398069230664e-05,
+      "loss": 0.9001799583435058,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46005,
+      "step": 930,
+      "tokens/total": 30474240,
+      "tokens/train_per_sec_per_gpu": 188.03,
+      "tokens/trainable": 28256136
+    },
+    {
+      "epoch": 0.6749237120804165,
+      "grad_norm": 0.355441153049469,
+      "learning_rate": 2.6175135319071025e-05,
+      "loss": 0.8904359817504883,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.43619,
+      "step": 940,
+      "tokens/total": 30801920,
+      "tokens/train_per_sec_per_gpu": 171.46,
+      "tokens/trainable": 28561112
+    },
+    {
+      "epoch": 0.6821037515706336,
+      "grad_norm": 0.33440473675727844,
+      "learning_rate": 2.513732642691107e-05,
+      "loss": 0.9151079177856445,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.49704,
+      "step": 950,
+      "tokens/total": 31129600,
+      "tokens/train_per_sec_per_gpu": 180.47,
+      "tokens/trainable": 28852180
+    },
+    {
+      "epoch": 0.6892837910608508,
+      "grad_norm": 0.34756094217300415,
+      "learning_rate": 2.4113557441555383e-05,
+      "loss": 0.9022106170654297,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46505,
+      "step": 960,
+      "tokens/total": 31457280,
+      "tokens/train_per_sec_per_gpu": 158.15,
+      "tokens/trainable": 29149028
+    },
+    {
+      "epoch": 0.6964638305510681,
+      "grad_norm": 0.33276310563087463,
+      "learning_rate": 2.310440648349859e-05,
+      "loss": 0.8890485763549805,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.43281,
+      "step": 970,
+      "tokens/total": 31784960,
+      "tokens/train_per_sec_per_gpu": 173.23,
+      "tokens/trainable": 29444158
+    },
+    {
+      "epoch": 0.7036438700412853,
+      "grad_norm": 0.3310735821723938,
+      "learning_rate": 2.2110443418461725e-05,
+      "loss": 0.9073732376098633,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.47781,
+      "step": 980,
+      "tokens/total": 32112640,
+      "tokens/train_per_sec_per_gpu": 192.02,
+      "tokens/trainable": 29737064
+    },
+    {
+      "epoch": 0.7108239095315024,
+      "grad_norm": 0.33712413907051086,
+      "learning_rate": 2.1132229535590093e-05,
+      "loss": 0.9036993026733399,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46872,
+      "step": 990,
+      "tokens/total": 32440320,
+      "tokens/train_per_sec_per_gpu": 187.3,
+      "tokens/trainable": 30032704
+    },
+    {
+      "epoch": 0.7180039490217196,
+      "grad_norm": 0.3424170911312103,
+      "learning_rate": 2.0170317230494316e-05,
+      "loss": 0.8963610649108886,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.45067,
+      "step": 1000,
+      "tokens/total": 32768000,
+      "tokens/train_per_sec_per_gpu": 168.64,
+      "tokens/trainable": 30328108
+    },
+    {
+      "epoch": 0.7251839885119368,
+      "grad_norm": 0.3354298174381256,
+      "learning_rate": 1.9225249693313548e-05,
+      "loss": 0.9032676696777344,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46765,
+      "step": 1010,
+      "tokens/total": 33095680,
+      "tokens/train_per_sec_per_gpu": 179.78,
+      "tokens/trainable": 30627696
+    },
+    {
+      "epoch": 0.732364028002154,
+      "grad_norm": 0.3396533727645874,
+      "learning_rate": 1.829756060197692e-05,
+      "loss": 0.8936196327209472,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.44396,
+      "step": 1020,
+      "tokens/total": 33423360,
+      "tokens/train_per_sec_per_gpu": 191.33,
+      "tokens/trainable": 30925488
+    },
+    {
+      "epoch": 0.7395440674923712,
+      "grad_norm": 0.3350459933280945,
+      "learning_rate": 1.7387773820836667e-05,
+      "loss": 0.9024556159973145,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46565,
+      "step": 1030,
+      "tokens/total": 33751040,
+      "tokens/train_per_sec_per_gpu": 189.71,
+      "tokens/trainable": 31223512
+    },
+    {
+      "epoch": 0.7467241069825884,
+      "grad_norm": 0.33155035972595215,
+      "learning_rate": 1.64964031048428e-05,
+      "loss": 0.8976601600646973,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.45385,
+      "step": 1040,
+      "tokens/total": 34078720,
+      "tokens/train_per_sec_per_gpu": 190.37,
+      "tokens/trainable": 31524356
+    },
+    {
+      "epoch": 0.7539041464728056,
+      "grad_norm": 0.34476324915885925,
+      "learning_rate": 1.5623951809426664e-05,
+      "loss": 0.9035608291625976,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46838,
+      "step": 1050,
+      "tokens/total": 34406400,
+      "tokens/train_per_sec_per_gpu": 173.32,
+      "tokens/trainable": 31817444
+    },
+    {
+      "epoch": 0.7610841859630227,
+      "grad_norm": 0.33920320868492126,
+      "learning_rate": 1.4770912606257004e-05,
+      "loss": 0.8911005020141601,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.43781,
+      "step": 1060,
+      "tokens/total": 34734080,
+      "tokens/train_per_sec_per_gpu": 165.81,
+      "tokens/trainable": 32115804
+    },
+    {
+      "epoch": 0.76826422545324,
+      "grad_norm": 0.3369329869747162,
+      "learning_rate": 1.3937767205029196e-05,
+      "loss": 0.8989034652709961,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.45691,
+      "step": 1070,
+      "tokens/total": 35061760,
+      "tokens/train_per_sec_per_gpu": 149.71,
+      "tokens/trainable": 32409324
+    },
+    {
+      "epoch": 0.7754442649434572,
+      "grad_norm": 0.3363012969493866,
+      "learning_rate": 1.3124986081444624e-05,
+      "loss": 0.9089801788330079,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.48179,
+      "step": 1080,
+      "tokens/total": 35389440,
+      "tokens/train_per_sec_per_gpu": 181.74,
+      "tokens/trainable": 32699570
+    },
+    {
+      "epoch": 0.7826243044336744,
+      "grad_norm": 0.34770333766937256,
+      "learning_rate": 1.2333028211533914e-05,
+      "loss": 0.9037967681884765,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46896,
+      "step": 1090,
+      "tokens/total": 35717120,
+      "tokens/train_per_sec_per_gpu": 187.48,
+      "tokens/trainable": 32993832
+    },
+    {
+      "epoch": 0.7898043439238915,
+      "grad_norm": 0.34793657064437866,
+      "learning_rate": 1.1562340812474004e-05,
+      "loss": 0.8945337295532226,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.44619,
+      "step": 1100,
+      "tokens/total": 36044800,
+      "tokens/train_per_sec_per_gpu": 162.93,
+      "tokens/trainable": 33296596
+    },
+    {
+      "epoch": 0.7969843834141088,
+      "grad_norm": 0.3449819087982178,
+      "learning_rate": 1.0813359090045411e-05,
+      "loss": 0.8774575233459473,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.40478,
+      "step": 1110,
+      "tokens/total": 36372480,
+      "tokens/train_per_sec_per_gpu": 163.01,
+      "tokens/trainable": 33597972
+    },
+    {
+      "epoch": 0.804164422904326,
+      "grad_norm": 0.35165634751319885,
+      "learning_rate": 1.0086505992872303e-05,
+      "loss": 0.8890524864196777,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.43282,
+      "step": 1120,
+      "tokens/total": 36700160,
+      "tokens/train_per_sec_per_gpu": 175.86,
+      "tokens/trainable": 33893216
+    },
+    {
+      "epoch": 0.8113444623945432,
+      "grad_norm": 0.34267014265060425,
+      "learning_rate": 9.382191973584193e-06,
+      "loss": 0.8928565979003906,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.4421,
+      "step": 1130,
+      "tokens/total": 37027840,
+      "tokens/train_per_sec_per_gpu": 154.7,
+      "tokens/trainable": 34185752
+    },
+    {
+      "epoch": 0.8185245018847603,
+      "grad_norm": 0.33240991830825806,
+      "learning_rate": 8.70081475703406e-06,
+      "loss": 0.8969085693359375,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.45201,
+      "step": 1140,
+      "tokens/total": 37355520,
+      "tokens/train_per_sec_per_gpu": 191.6,
+      "tokens/trainable": 34477912
+    },
+    {
+      "epoch": 0.8257045413749775,
+      "grad_norm": 0.3527986407279968,
+      "learning_rate": 8.04275911570389e-06,
+      "loss": 0.9027474403381348,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46637,
+      "step": 1150,
+      "tokens/total": 37683200,
+      "tokens/train_per_sec_per_gpu": 190.98,
+      "tokens/trainable": 34776324
+    },
+    {
+      "epoch": 0.8328845808651948,
+      "grad_norm": 0.33955442905426025,
+      "learning_rate": 7.4083966524242706e-06,
+      "loss": 0.8867701530456543,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.42728,
+      "step": 1160,
+      "tokens/total": 38010880,
+      "tokens/train_per_sec_per_gpu": 152.15,
+      "tokens/trainable": 35068108
+    },
+    {
+      "epoch": 0.840064620355412,
+      "grad_norm": 0.35017237067222595,
+      "learning_rate": 6.7980855905310115e-06,
+      "loss": 0.8830046653747559,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.41815,
+      "step": 1170,
+      "tokens/total": 38338560,
+      "tokens/train_per_sec_per_gpu": 158.17,
+      "tokens/trainable": 35365276
+    },
+    {
+      "epoch": 0.8472446598456291,
+      "grad_norm": 0.3360691964626312,
+      "learning_rate": 6.212170571577086e-06,
+      "loss": 0.9050349235534668,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.47202,
+      "step": 1180,
+      "tokens/total": 38666240,
+      "tokens/train_per_sec_per_gpu": 188.59,
+      "tokens/trainable": 35662984
+    },
+    {
+      "epoch": 0.8544246993358463,
+      "grad_norm": 0.35091644525527954,
+      "learning_rate": 5.650982460714083e-06,
+      "loss": 0.8928234100341796,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.44201,
+      "step": 1190,
+      "tokens/total": 38993920,
+      "tokens/train_per_sec_per_gpu": 174.45,
+      "tokens/trainable": 35961584
+    },
+    {
+      "epoch": 0.8616047388260636,
+      "grad_norm": 0.3266690969467163,
+      "learning_rate": 5.1148381598533355e-06,
+      "loss": 0.8990215301513672,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.4572,
+      "step": 1200,
+      "tokens/total": 39321600,
+      "tokens/train_per_sec_per_gpu": 168.61,
+      "tokens/trainable": 36257436
+    },
+    {
+      "epoch": 0.8687847783162808,
+      "grad_norm": 0.34352022409439087,
+      "learning_rate": 4.604040428711992e-06,
+      "loss": 0.8924401283264161,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.44108,
+      "step": 1210,
+      "tokens/total": 39649280,
+      "tokens/train_per_sec_per_gpu": 159.43,
+      "tokens/trainable": 36553308
+    },
+    {
+      "epoch": 0.8759648178064979,
+      "grad_norm": 0.3505189120769501,
+      "learning_rate": 4.118877713845049e-06,
+      "loss": 0.8806687355041504,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.41251,
+      "step": 1220,
+      "tokens/total": 39976960,
+      "tokens/train_per_sec_per_gpu": 167.75,
+      "tokens/trainable": 36845968
+    },
+    {
+      "epoch": 0.8831448572967151,
+      "grad_norm": 0.3569996654987335,
+      "learning_rate": 3.659623985760213e-06,
+      "loss": 0.9029593467712402,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46689,
+      "step": 1230,
+      "tokens/total": 40304640,
+      "tokens/train_per_sec_per_gpu": 162.27,
+      "tokens/trainable": 37137272
+    },
+    {
+      "epoch": 0.8903248967869323,
+      "grad_norm": 0.3513758182525635,
+      "learning_rate": 3.226538584207228e-06,
+      "loss": 0.8935052871704101,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.44368,
+      "step": 1240,
+      "tokens/total": 40632320,
+      "tokens/train_per_sec_per_gpu": 178.14,
+      "tokens/trainable": 37433968
+    },
+    {
+      "epoch": 0.8975049362771496,
+      "grad_norm": 0.3621438443660736,
+      "learning_rate": 2.819866071729127e-06,
+      "loss": 0.9010676383972168,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.46223,
+      "step": 1250,
+      "tokens/total": 40960000,
+      "tokens/train_per_sec_per_gpu": 151.39,
+      "tokens/trainable": 37727884
+    },
+    {
+      "epoch": 0.9046849757673667,
+      "grad_norm": 0.34137412905693054,
+      "learning_rate": 2.439836095558262e-06,
+      "loss": 0.8975703239440918,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.45363,
+      "step": 1260,
+      "tokens/total": 41287680,
+      "tokens/train_per_sec_per_gpu": 177.06,
+      "tokens/trainable": 38024668
+    },
+    {
+      "epoch": 0.9118650152575839,
+      "grad_norm": 0.33366984128952026,
+      "learning_rate": 2.08666325793494e-06,
+      "loss": 0.8833458900451661,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.41898,
+      "step": 1270,
+      "tokens/total": 41615360,
+      "tokens/train_per_sec_per_gpu": 181.67,
+      "tokens/trainable": 38318496
+    },
+    {
+      "epoch": 0.9190450547478011,
+      "grad_norm": 0.3441231846809387,
+      "learning_rate": 1.7605469949218579e-06,
+      "loss": 0.8940237998962403,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.44495,
+      "step": 1280,
+      "tokens/total": 41943040,
+      "tokens/train_per_sec_per_gpu": 190.27,
+      "tokens/trainable": 38611004
+    },
+    {
+      "epoch": 0.9262250942380184,
+      "grad_norm": 0.3395031690597534,
+      "learning_rate": 1.461671463782982e-06,
+      "loss": 0.8930609703063965,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.44259,
+      "step": 1290,
+      "tokens/total": 42270720,
+      "tokens/train_per_sec_per_gpu": 193.08,
+      "tokens/trainable": 38907484
+    },
+    {
+      "epoch": 0.9334051337282355,
+      "grad_norm": 0.35241463780403137,
+      "learning_rate": 1.190205438990266e-06,
+      "loss": 0.896492862701416,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.45099,
+      "step": 1300,
+      "tokens/total": 42598400,
+      "tokens/train_per_sec_per_gpu": 156.86,
+      "tokens/trainable": 39203216
+    },
+    {
+      "epoch": 0.9405851732184527,
+      "grad_norm": 0.33634573221206665,
+      "learning_rate": 9.463022169169667e-07,
+      "loss": 0.8888393402099609,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.4323,
+      "step": 1310,
+      "tokens/total": 42926080,
+      "tokens/train_per_sec_per_gpu": 190.41,
+      "tokens/trainable": 39501840
+    },
+    {
+      "epoch": 0.9477652127086699,
+      "grad_norm": 0.3354337513446808,
+      "learning_rate": 7.300995292715107e-07,
+      "loss": 0.884650707244873,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.42214,
+      "step": 1320,
+      "tokens/total": 43253760,
+      "tokens/train_per_sec_per_gpu": 189.56,
+      "tokens/trainable": 39792848
+    },
+    {
+      "epoch": 0.954945252198887,
+      "grad_norm": 0.3369513750076294,
+      "learning_rate": 5.417194653206337e-07,
+      "loss": 0.8907259941101074,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.4369,
+      "step": 1330,
+      "tokens/total": 43581440,
+      "tokens/train_per_sec_per_gpu": 163.01,
+      "tokens/trainable": 40079552
+    },
+    {
+      "epoch": 0.9621252916891043,
+      "grad_norm": 0.3338170349597931,
+      "learning_rate": 3.8126840294576136e-07,
+      "loss": 0.8961419105529785,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.45013,
+      "step": 1340,
+      "tokens/total": 43909120,
+      "tokens/train_per_sec_per_gpu": 187.9,
+      "tokens/trainable": 40373720
+    },
+    {
+      "epoch": 0.9693053311793215,
+      "grad_norm": 0.348407506942749,
+      "learning_rate": 2.488369485716513e-07,
+      "loss": 0.909212303161621,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.48237,
+      "step": 1350,
+      "tokens/total": 44236800,
+      "tokens/train_per_sec_per_gpu": 164.03,
+      "tokens/trainable": 40660548
+    },
+    {
+      "epoch": 0.9764853706695387,
+      "grad_norm": 0.35887008905410767,
+      "learning_rate": 1.4449988600111487e-07,
+      "loss": 0.8935896873474121,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.44389,
+      "step": 1360,
+      "tokens/total": 44564480,
+      "tokens/train_per_sec_per_gpu": 148.51,
+      "tokens/trainable": 40953568
+    },
+    {
+      "epoch": 0.9836654101597558,
+      "grad_norm": 0.3604047894477844,
+      "learning_rate": 6.831613418468163e-08,
+      "loss": 0.8940872192382813,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.4451,
+      "step": 1370,
+      "tokens/total": 44892160,
+      "tokens/train_per_sec_per_gpu": 167.57,
+      "tokens/trainable": 41241024
+    },
+    {
+      "epoch": 0.9908454496499731,
+      "grad_norm": 0.3480483889579773,
+      "learning_rate": 2.0328713949230305e-08,
+      "loss": 0.8837766647338867,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.42002,
+      "step": 1380,
+      "tokens/total": 45219840,
+      "tokens/train_per_sec_per_gpu": 189.59,
+      "tokens/trainable": 41533044
+    },
+    {
+      "epoch": 0.9980254891401903,
+      "grad_norm": 0.3344267010688782,
+      "learning_rate": 5.647237041939146e-10,
+      "loss": 0.8873067855834961,
+      "memory/device_reserved (GiB)": 48.31,
+      "memory/max_active (GiB)": 44.19,
+      "memory/max_allocated (GiB)": 44.19,
+      "ppl": 2.42858,
+      "step": 1390,
+      "tokens/total": 45547520,
+      "tokens/train_per_sec_per_gpu": 189.55,
+      "tokens/trainable": 41824232
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1391,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 696,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.3263539281327555e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/config.json

@@ -0,0 +1,110 @@
+{
+  "architectures": [
+    "Qwen3_5ForConditionalGeneration"
+  ],
+  "dtype": "bfloat16",
+  "image_token_id": 248056,
+  "model_type": "qwen3_5",
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attn_output_gate": true,
+    "bos_token_id": null,
+    "dtype": "bfloat16",
+    "eos_token_id": 248044,
+    "full_attention_interval": 4,
+    "head_dim": 256,
+    "hidden_act": "silu",
+    "hidden_size": 4096,
+    "initializer_range": 0.02,
+    "intermediate_size": 12288,
+    "layer_types": [
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention",
+      "linear_attention",
+      "linear_attention",
+      "linear_attention",
+      "full_attention"
+    ],
+    "linear_conv_kernel_dim": 4,
+    "linear_key_head_dim": 128,
+    "linear_num_key_heads": 16,
+    "linear_num_value_heads": 32,
+    "linear_value_head_dim": 128,
+    "mamba_ssm_dtype": "float32",
+    "max_position_embeddings": 262144,
+    "mlp_only_layers": [],
+    "model_type": "qwen3_5_text",
+    "mtp_num_hidden_layers": 1,
+    "mtp_use_dedicated_embeddings": false,
+    "num_attention_heads": 16,
+    "num_hidden_layers": 32,
+    "num_key_value_heads": 4,
+    "pad_token_id": null,
+    "partial_rotary_factor": 0.25,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "mrope_interleaved": true,
+      "mrope_section": [
+        11,
+        11,
+        10
+      ],
+      "partial_rotary_factor": 0.25,
+      "rope_theta": 10000000,
+      "rope_type": "default"
+    },
+    "tie_word_embeddings": false,
+    "use_cache": false,
+    "vocab_size": 248320
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.12.0",
+  "use_cache": false,
+  "video_token_id": 248057,
+  "vision_config": {
+    "deepstack_visual_indexes": [],
+    "depth": 27,
+    "dtype": "bfloat16",
+    "hidden_act": "gelu_pytorch_tanh",
+    "hidden_size": 1152,
+    "in_channels": 3,
+    "initializer_range": 0.02,
+    "intermediate_size": 4304,
+    "model_type": "qwen3_5_vision",
+    "num_heads": 16,
+    "num_position_embeddings": 2304,
+    "out_hidden_size": 4096,
+    "patch_size": 16,
+    "spatial_merge_size": 2,
+    "temporal_patch_size": 2
+  },
+  "vision_end_token_id": 248054,
+  "vision_start_token_id": 248053
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/debug.log

@@ -0,0 +1,397 @@
+[2026-06-19 09:36:57,920] [DEBUG] [axolotl.utils.config.log_gpu_memory_usage:127] [PID:75323] baseline 0.000GB ()
+[2026-06-19 09:36:57,921] [INFO] [axolotl.cli.config.load_cfg:333] [PID:75323] config:
+{
+  "activation_offloading": false,
+  "adapter": "lora",
+  "attn_implementation": "flash_attention_2",
+  "attn_needs_dtype_cast": true,
+  "attn_supports_packing": true,
+  "attn_uses_flash_lib": true,
+  "auto_resume_from_checkpoints": true,
+  "axolotl_config_path": "/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/axolotl/msm.yaml",
+  "base_model": "Qwen/Qwen3.5-9B",
+  "base_model_config": "Qwen/Qwen3.5-9B",
+  "batch_size": 4,
+  "bf16": true,
+  "capabilities": {
+    "bf16": true,
+    "compute_capability": "sm_90",
+    "fp8": true,
+    "n_gpu": 1,
+    "n_node": 1,
+    "tf32": true
+  },
+  "chat_template": "tokenizer_default",
+  "context_parallel_size": 1,
+  "dataloader_num_workers": 1,
+  "dataloader_pin_memory": true,
+  "dataloader_prefetch_factor": 256,
+  "dataset_num_proc": 32,
+  "dataset_prepared_path": "/workspace/mats_project/data/.axolotl-prepared-cache",
+  "datasets": [
+    {
+      "field": "text",
+      "message_property_mappings": {
+        "content": "content",
+        "role": "role"
+      },
+      "path": "/workspace/mats_project/data/msm/msm-qwen-philosophy-spec.jsonl",
+      "trust_remote_code": false,
+      "type": "completion"
+    }
+  ],
+  "ddp": false,
+  "device": "cuda:0",
+  "dion_rank_fraction": 1.0,
+  "dion_rank_multiple_of": 1,
+  "eaft_alpha": 1.0,
+  "eaft_k": 20,
+  "env_capabilities": {
+    "torch_version": "2.12.1"
+  },
+  "eval_batch_size": 1,
+  "eval_causal_lm_metrics": [
+    "sacrebleu",
+    "comet",
+    "ter",
+    "chrf"
+  ],
+  "eval_max_new_tokens": 128,
+  "eval_sample_packing": true,
+  "eval_table_size": 0,
+  "experimental_skip_move_to_device": true,
+  "fp16": false,
+  "generate_samples": false,
+  "generation_do_sample": true,
+  "generation_max_new_tokens": 50,
+  "generation_prompt_ratio": 0.5,
+  "generation_temperature": 0.7,
+  "gradient_accumulation_steps": 4,
+  "gradient_checkpointing": true,
+  "gradient_checkpointing_kwargs": {
+    "use_reentrant": true
+  },
+  "include_tkps": true,
+  "is_multimodal": true,
+  "layer_offloading": false,
+  "learning_rate": 0.0001,
+  "lisa_layers_attribute": "model.layers",
+  "load_best_model_at_end": false,
+  "load_in_4bit": false,
+  "load_in_8bit": false,
+  "local_rank": 0,
+  "logging_steps": 10,
+  "lora_alpha": 128,
+  "lora_dropout": 0.0,
+  "lora_mlp_kernel": true,
+  "lora_o_kernel": true,
+  "lora_qkv_kernel": true,
+  "lora_r": 64,
+  "lora_target_modules": [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "up_proj",
+    "down_proj"
+  ],
+  "loraplus_lr_embedding": 1e-06,
+  "lr_scheduler": "cosine",
+  "max_grad_norm": 1.0,
+  "mean_resizing_embeddings": false,
+  "merge_method": "memory_efficient",
+  "micro_batch_size": 1,
+  "model_config_type": "qwen3_5",
+  "model_config_type_text": "qwen3_5_text",
+  "num_epochs": 1.0,
+  "num_generation_samples": 3,
+  "optimizer": "adamw_torch_fused",
+  "otel_metrics_host": "localhost",
+  "otel_metrics_port": 8000,
+  "output_dir": "/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm",
+  "pad_to_sequence_len": true,
+  "pretrain_multipack_attn": true,
+  "processor_config": "Qwen/Qwen3.5-9B",
+  "profiler_steps_start": 0,
+  "qgalore_cos_threshold": 0.4,
+  "qgalore_gamma_proj": 2,
+  "qgalore_proj_bits": 4,
+  "qgalore_proj_group_size": 256,
+  "qgalore_proj_quant": true,
+  "qgalore_proj_type": "std",
+  "qgalore_queue_size": 5,
+  "qgalore_rank": 256,
+  "qgalore_scale": 0.25,
+  "qgalore_update_proj_gap": 200,
+  "qlora_sharded_model_loading": false,
+  "quantize_moe_experts": false,
+  "ray_num_workers": 1,
+  "relora_prune_method": "magnitude",
+  "resources_per_worker": {
+    "GPU": 1
+  },
+  "sample_packing": true,
+  "sample_packing_bin_size": 200,
+  "sample_packing_group_size": 100000,
+  "save_only_model": true,
+  "save_safetensors": true,
+  "save_steps": 0.5,
+  "save_total_limit": 1,
+  "saves_per_epoch": 2,
+  "sequence_len": 8192,
+  "shuffle_before_merging_datasets": false,
+  "shuffle_merged_datasets": true,
+  "skip_prepare_dataset": false,
+  "special_tokens": {
+    "eos_token": "<|im_end|>",
+    "pad_token": "<|endoftext|>"
+  },
+  "streaming_multipack_buffer_size": 10000,
+  "strict": false,
+  "tensor_parallel_size": 1,
+  "tf32": true,
+  "tiled_mlp_use_original_mlp": true,
+  "tokenizer_config": "Qwen/Qwen3.5-9B",
+  "tokenizer_save_jinja_files": true,
+  "torch_dtype": "torch.bfloat16",
+  "train_on_inputs": false,
+  "trl": {
+    "async_prefetch": false,
+    "log_completions": false,
+    "mask_truncated_completions": false,
+    "ref_model_mixup_alpha": 0.9,
+    "ref_model_sync_steps": 64,
+    "replay_buffer_size": 0,
+    "replay_recompute_logps": true,
+    "reroll_max_groups": 1,
+    "reroll_start_fraction": 1.0,
+    "reward_num_workers": 1,
+    "scale_rewards": true,
+    "skip_zero_advantage_batches": true,
+    "sync_ref_model": false,
+    "use_data_producer": false,
+    "use_vllm": false,
+    "vllm_lora_sync": false,
+    "vllm_server_host": "0.0.0.0",
+    "vllm_server_port": 8000
+  },
+  "use_otel_metrics": false,
+  "use_ray": false,
+  "use_wandb": true,
+  "val_set_size": 0.0,
+  "vllm": {
+    "device": "auto",
+    "dtype": "auto",
+    "gpu_memory_utilization": 0.9,
+    "host": "0.0.0.0",
+    "port": 8000
+  },
+  "wandb_name": "philosophy-msm-aft-instruct-20260619-093540/msm",
+  "wandb_project": "why-gen",
+  "warmup_ratio": 0.05,
+  "weight_decay": 0.01,
+  "world_size": 1
+}
+[2026-06-19 09:36:59,248] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:75323] EOS: 248046 / <|im_end|>
+[2026-06-19 09:36:59,248] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:75323] BOS: None / None
+[2026-06-19 09:36:59,248] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:75323] PAD: 248044 / <|endoftext|>
+[2026-06-19 09:36:59,248] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:75323] UNK: None / None
+[2026-06-19 09:36:59,259] [INFO] [axolotl.utils.data.shared.load_preprocessed_dataset:477] [PID:75323] Loading prepared dataset from disk at /workspace/mats_project/data/.axolotl-prepared-cache/5ebaa067467de1cd17f91a288757e1fa...
+
+[2026-06-19 09:36:59,437] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:420] [PID:75323] total_num_tokens: 41_882_726
+[2026-06-19 09:36:59,660] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:438] [PID:75323] `total_supervised_tokens: 41_882_726`
+[2026-06-19 09:37:01,777] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:75323] generate_batches time: 0.9359323978424072
+[2026-06-19 09:37:02,786] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:75323] generate_batches time: 1.007965326309204
+[2026-06-19 09:37:03,884] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:75323] generate_batches time: 1.0978868007659912
+[2026-06-19 09:37:04,809] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:75323] generate_batches time: 0.9238288402557373
+[2026-06-19 09:37:04,840] [INFO] [axolotl.utils.samplers.multipack.calc_min_len:438] [PID:75323] gather_len_batches: [5567]
+[2026-06-19 09:37:04,840] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:495] [PID:75323] data_loader_len: 1391
+[2026-06-19 09:37:04,840] [INFO] [axolotl.utils.trainer.calc_sample_packing_eff_est:504] [PID:75323] sample_packing_eff_est across ranks: [0.9188780465801357]
+[2026-06-19 09:37:04,840] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:516] [PID:75323] sample_packing_eff_est: 0.92
+[2026-06-19 09:37:04,840] [DEBUG] [axolotl.utils.trainer.calculate_total_num_steps:521] [PID:75323] total_num_steps: 1391
+[2026-06-19 09:37:04,841] [INFO] [axolotl.utils.data.sft._prepare_standard_dataset:121] [PID:75323] Maximum number of steps set at 1391
+[2026-06-19 09:37:04,904] [DEBUG] [axolotl.train.setup_model_and_tokenizer:70] [PID:75323] loading tokenizer... Qwen/Qwen3.5-9B
+[2026-06-19 09:37:06,100] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:311] [PID:75323] EOS: 248046 / <|im_end|>
+[2026-06-19 09:37:06,100] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:312] [PID:75323] BOS: None / None
+[2026-06-19 09:37:06,100] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:313] [PID:75323] PAD: 248044 / <|endoftext|>
+[2026-06-19 09:37:06,100] [DEBUG] [axolotl.loaders.tokenizer.load_tokenizer:314] [PID:75323] UNK: None / None
+[2026-06-19 09:37:09,599] [DEBUG] [axolotl.train.setup_model_and_tokenizer:81] [PID:75323] Loading model
+[2026-06-19 09:37:09,710] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:75] [PID:75323] Patched OptimState8bit for torch.compile compatibility
+[2026-06-19 09:37:09,710] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:122] [PID:75323] Patched OptimState4bit for torch.compile compatibility
+[2026-06-19 09:37:09,710] [DEBUG] [axolotl.monkeypatch.torchao_optim.patch_torchao_optim_state_8bit:154] [PID:75323] Patched OptimStateFp8 for torch.compile compatibility
+[2026-06-19 09:37:09,723] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_evaluation_loop:94] [PID:75323] Patched Trainer.evaluation_loop with nanmean loss calculation
+[2026-06-19 09:37:09,724] [DEBUG] [axolotl.monkeypatch.transformers.trainer_loss_calc.patch_maybe_log_save_evaluate:148] [PID:75323] Patched Trainer._maybe_log_save_evaluate with nanmean loss calculation
+[2026-06-19 09:37:09,731] [INFO] [axolotl.monkeypatch.attention.flash_attn_4.patch_flash_attn_4:52] [PID:75323] Flash Attention 4 is available for your GPU and offers faster training speeds. To enable: pip install flash-attn-4
+[2026-06-19 09:37:13,253] [INFO] [axolotl.monkeypatch.lora_kernels.patch_self_attn_lora:304] [PID:75323] Patched attention class with LoRA optims: Qwen3_5Attention
+[2026-06-19 09:37:13,266] [INFO] [axolotl.monkeypatch.models.qwen3_5.modeling._apply_packing_patches:254] [PID:75323] Applied Qwen3_5 packing patch (fla_causal_conv1d=available)
+[2026-06-19 09:37:13,266] [INFO] [axolotl.monkeypatch.models.qwen3_5.modeling.patch_qwen3_5_vlm_flash_attention:289] [PID:75323] Applied Qwen3.5 VLM flash-attention patch (3-D MRoPE position_ids)
+[2026-06-19 09:37:13,269] [INFO] [axolotl.loaders.patch_manager._apply_multipack_patches:704] [PID:75323] Applying multipack dataloader patch for sample packing...
+
+
+
+[2026-06-19 09:37:17,288] [INFO] [axolotl.loaders.model._configure_embedding_dtypes:433] [PID:75323] Converting modules to torch.bfloat16
+[2026-06-19 09:37:19,642] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:75323] Memory usage after model load 0.000GB (+0.000GB allocated, +0.002GB reserved)
+trainable params: 116,391,936 || all params: 9,526,205,680 || trainable%: 1.2218
+[2026-06-19 09:37:20,462] [DEBUG] [axolotl.loaders.model.log_gpu_memory_usage:127] [PID:75323] after adapters 0.000GB ()
+[2026-06-19 09:37:24,654] [INFO] [axolotl.train.save_initial_configs:450] [PID:75323] Pre-saving adapter config to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm...
+[2026-06-19 09:37:24,658] [INFO] [axolotl.train.save_initial_configs:454] [PID:75323] Pre-saving tokenizer to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm...
+[2026-06-19 09:37:24,807] [INFO] [axolotl.train.save_initial_configs:459] [PID:75323] Pre-saving model config to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm...
+[2026-06-19 09:37:24,817] [INFO] [axolotl.train.save_initial_configs:463] [PID:75323] Pre-saving processor to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm...
+[2026-06-19 09:37:24,970] [INFO] [axolotl.train.execute_training:226] [PID:75323] Starting trainer...
+[2026-06-19 09:37:27,477] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:75323] generate_batches time: 1.0457372665405273
+[2026-06-19 09:37:28,550] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:75323] generate_batches time: 1.0726072788238525
+[2026-06-19 09:37:29,610] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:75323] generate_batches time: 1.0590167045593262
+[2026-06-19 09:37:30,721] [DEBUG] [axolotl.utils.samplers.multipack.__len__:462] [PID:75323] generate_batches time: 1.1108381748199463
+[2026-06-19 09:37:30,721] [INFO] [axolotl.utils.samplers.multipack.calc_min_len:438] [PID:75323] gather_len_batches: [5571]
+wandb: Tracking run with wandb version 0.27.2
+wandb: W&B syncing is set to `offline` in this directory. Run `wandb online` or set WANDB_MODE=online to enable cloud syncing.
+wandb: Run data is saved locally in /workspace/wandb/wandb/offline-run-20260619_093731-cupn5mww
+wandb: WARNING Saving files without folders. If you want to preserve subdirectories pass base_path to wandb.save, i.e. wandb.save("/mnt/folder/file.h5", base_path="/mnt")
+wandb: WARNING Symlinked 1 file into the W&B run directory; call wandb.save again to sync new files.
+[2026-06-19 09:37:32,632] [INFO] [axolotl.utils.callbacks.on_train_begin:807] [PID:75323] The Axolotl config has been saved to the WandB run under files.
+{'loss': '1.524', 'grad_norm': '0.5396', 'learning_rate': '1.304e-05', 'ppl': '4.591', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1874', 'tokens/total': 327680, 'tokens/trainable': 320233, 'epoch': '0.00718'}
+{'loss': '1.423', 'grad_norm': '0.3517', 'learning_rate': '2.754e-05', 'ppl': '4.15', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1916', 'tokens/total': 655360, 'tokens/trainable': 636503, 'epoch': '0.01436'}
+{'loss': '1.356', 'grad_norm': '0.3654', 'learning_rate': '4.203e-05', 'ppl': '3.879', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1730', 'tokens/total': 983040, 'tokens/trainable': 951070, 'epoch': '0.02154'}
+{'loss': '1.282', 'grad_norm': '0.3798', 'learning_rate': '5.652e-05', 'ppl': '3.605', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1853', 'tokens/total': 1310720, 'tokens/trainable': 1267181, 'epoch': '0.02872'}
+{'loss': '1.221', 'grad_norm': '0.4241', 'learning_rate': '7.101e-05', 'ppl': '3.39', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1915', 'tokens/total': 1638400, 'tokens/trainable': 1584678, 'epoch': '0.0359'}
+{'loss': '1.177', 'grad_norm': '0.4011', 'learning_rate': '8.551e-05', 'ppl': '3.246', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1911', 'tokens/total': 1966080, 'tokens/trainable': 1895876, 'epoch': '0.04308'}
+{'loss': '1.135', 'grad_norm': '0.3944', 'learning_rate': '0.0001', 'ppl': '3.112', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1710', 'tokens/total': 2293760, 'tokens/trainable': 2211142, 'epoch': '0.05026'}
+{'loss': '1.128', 'grad_norm': '0.4057', 'learning_rate': '9.999e-05', 'ppl': '3.089', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1920', 'tokens/total': 2621440, 'tokens/trainable': 2523632, 'epoch': '0.05744'}
+{'loss': '1.102', 'grad_norm': '0.349', 'learning_rate': '9.994e-05', 'ppl': '3.009', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1774', 'tokens/total': 2949120, 'tokens/trainable': 2835508, 'epoch': '0.06462'}
+{'loss': '1.095', 'grad_norm': '0.3582', 'learning_rate': '9.987e-05', 'ppl': '2.988', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1920', 'tokens/total': 3276800, 'tokens/trainable': 3148695, 'epoch': '0.0718'}
+{'loss': '1.075', 'grad_norm': '0.3543', 'learning_rate': '9.977e-05', 'ppl': '2.93', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1842', 'tokens/total': 3604480, 'tokens/trainable': 3466044, 'epoch': '0.07898'}
+{'loss': '1.068', 'grad_norm': '0.3599', 'learning_rate': '9.965e-05', 'ppl': '2.909', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1905', 'tokens/total': 3932160, 'tokens/trainable': 3778534, 'epoch': '0.08616'}
+{'loss': '1.056', 'grad_norm': '0.3381', 'learning_rate': '9.949e-05', 'ppl': '2.874', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1802', 'tokens/total': 4259840, 'tokens/trainable': 4091431, 'epoch': '0.09334'}
+{'loss': '1.05', 'grad_norm': '0.3409', 'learning_rate': '9.931e-05', 'ppl': '2.859', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1901', 'tokens/total': 4587520, 'tokens/trainable': 4401406, 'epoch': '0.1005'}
+{'loss': '1.048', 'grad_norm': '0.3436', 'learning_rate': '9.91e-05', 'ppl': '2.853', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1892', 'tokens/total': 4915200, 'tokens/trainable': 4712142, 'epoch': '0.1077'}
+{'loss': '1.042', 'grad_norm': '0.3363', 'learning_rate': '9.886e-05', 'ppl': '2.834', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1698', 'tokens/total': 5242880, 'tokens/trainable': 5025694, 'epoch': '0.1149'}
+{'loss': '1.033', 'grad_norm': '0.3273', 'learning_rate': '9.859e-05', 'ppl': '2.811', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1721', 'tokens/total': 5570560, 'tokens/trainable': 5333753, 'epoch': '0.1221'}
+{'loss': '1.031', 'grad_norm': '0.344', 'learning_rate': '9.83e-05', 'ppl': '2.804', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1892', 'tokens/total': 5898240, 'tokens/trainable': 5641801, 'epoch': '0.1292'}
+{'loss': '1.011', 'grad_norm': '0.3328', 'learning_rate': '9.798e-05', 'ppl': '2.749', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1751', 'tokens/total': 6225920, 'tokens/trainable': 5951459, 'epoch': '0.1364'}
+{'loss': '1.015', 'grad_norm': '0.3229', 'learning_rate': '9.763e-05', 'ppl': '2.759', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1911', 'tokens/total': 6553600, 'tokens/trainable': 6261411, 'epoch': '0.1436'}
+{'loss': '1.011', 'grad_norm': '0.3312', 'learning_rate': '9.726e-05', 'ppl': '2.749', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1916', 'tokens/total': 6881280, 'tokens/trainable': 6565760, 'epoch': '0.1508'}
+{'loss': '1.015', 'grad_norm': '0.3399', 'learning_rate': '9.686e-05', 'ppl': '2.76', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1921', 'tokens/total': 7208960, 'tokens/trainable': 6876578, 'epoch': '0.158'}
+{'loss': '0.9883', 'grad_norm': '0.3277', 'learning_rate': '9.643e-05', 'ppl': '2.687', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1612', 'tokens/total': 7536640, 'tokens/trainable': 7181735, 'epoch': '0.1651'}
+{'loss': '1.014', 'grad_norm': '0.3264', 'learning_rate': '9.598e-05', 'ppl': '2.756', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1672', 'tokens/total': 7864320, 'tokens/trainable': 7490134, 'epoch': '0.1723'}
+{'loss': '0.9791', 'grad_norm': '0.3332', 'learning_rate': '9.55e-05', 'ppl': '2.662', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1664', 'tokens/total': 8192000, 'tokens/trainable': 7794665, 'epoch': '0.1795'}
+{'loss': '1.002', 'grad_norm': '0.3295', 'learning_rate': '9.499e-05', 'ppl': '2.724', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1645', 'tokens/total': 8519680, 'tokens/trainable': 8094467, 'epoch': '0.1867'}
+{'loss': '0.9751', 'grad_norm': '0.3087', 'learning_rate': '9.446e-05', 'ppl': '2.651', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1841', 'tokens/total': 8847360, 'tokens/trainable': 8395919, 'epoch': '0.1939'}
+{'loss': '0.9886', 'grad_norm': '0.3183', 'learning_rate': '9.39e-05', 'ppl': '2.688', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1571', 'tokens/total': 9175040, 'tokens/trainable': 8701190, 'epoch': '0.201'}
+{'loss': '0.9843', 'grad_norm': '0.321', 'learning_rate': '9.332e-05', 'ppl': '2.676', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1924', 'tokens/total': 9502720, 'tokens/trainable': 9007349, 'epoch': '0.2082'}
+{'loss': '0.9741', 'grad_norm': '0.3173', 'learning_rate': '9.272e-05', 'ppl': '2.649', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1909', 'tokens/total': 9830400, 'tokens/trainable': 9310722, 'epoch': '0.2154'}
+{'loss': '0.9714', 'grad_norm': '0.3275', 'learning_rate': '9.209e-05', 'ppl': '2.642', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1861', 'tokens/total': 10158080, 'tokens/trainable': 9620736, 'epoch': '0.2226'}
+{'loss': '0.9776', 'grad_norm': '0.3386', 'learning_rate': '9.143e-05', 'ppl': '2.658', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1906', 'tokens/total': 10485760, 'tokens/trainable': 9926004, 'epoch': '0.2298'}
+{'loss': '0.9853', 'grad_norm': '0.3339', 'learning_rate': '9.076e-05', 'ppl': '2.679', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1918', 'tokens/total': 10813440, 'tokens/trainable': 10230832, 'epoch': '0.2369'}
+{'loss': '0.9702', 'grad_norm': '0.3052', 'learning_rate': '9.006e-05', 'ppl': '2.638', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1902', 'tokens/total': 11141120, 'tokens/trainable': 10530921, 'epoch': '0.2441'}
+{'loss': '0.9703', 'grad_norm': '0.319', 'learning_rate': '8.933e-05', 'ppl': '2.639', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1793', 'tokens/total': 11468800, 'tokens/trainable': 10835657, 'epoch': '0.2513'}
+{'loss': '0.9586', 'grad_norm': '0.3331', 'learning_rate': '8.859e-05', 'ppl': '2.608', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1655', 'tokens/total': 11796480, 'tokens/trainable': 11132041, 'epoch': '0.2585'}
+{'loss': '0.9707', 'grad_norm': '0.3268', 'learning_rate': '8.782e-05', 'ppl': '2.64', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1888', 'tokens/total': 12124160, 'tokens/trainable': 11432692, 'epoch': '0.2657'}
+{'loss': '0.9769', 'grad_norm': '0.3501', 'learning_rate': '8.704e-05', 'ppl': '2.656', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1808', 'tokens/total': 12451840, 'tokens/trainable': 11734365, 'epoch': '0.2728'}
+{'loss': '0.9685', 'grad_norm': '0.3342', 'learning_rate': '8.623e-05', 'ppl': '2.634', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1657', 'tokens/total': 12779520, 'tokens/trainable': 12036905, 'epoch': '0.28'}
+{'loss': '0.9672', 'grad_norm': '0.3277', 'learning_rate': '8.54e-05', 'ppl': '2.63', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1541', 'tokens/total': 13107200, 'tokens/trainable': 12342375, 'epoch': '0.2872'}
+{'loss': '0.9673', 'grad_norm': '0.3244', 'learning_rate': '8.455e-05', 'ppl': '2.631', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1917', 'tokens/total': 13434880, 'tokens/trainable': 12642230, 'epoch': '0.2944'}
+{'loss': '0.968', 'grad_norm': '0.3343', 'learning_rate': '8.368e-05', 'ppl': '2.633', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1883', 'tokens/total': 13762560, 'tokens/trainable': 12943390, 'epoch': '0.3016'}
+{'loss': '0.9359', 'grad_norm': '0.3149', 'learning_rate': '8.279e-05', 'ppl': '2.55', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1923', 'tokens/total': 14090240, 'tokens/trainable': 13246463, 'epoch': '0.3087'}
+{'loss': '0.9469', 'grad_norm': '0.3205', 'learning_rate': '8.189e-05', 'ppl': '2.578', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1749', 'tokens/total': 14417920, 'tokens/trainable': 13545778, 'epoch': '0.3159'}
+{'loss': '0.9461', 'grad_norm': '0.3257', 'learning_rate': '8.096e-05', 'ppl': '2.576', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1865', 'tokens/total': 14745600, 'tokens/trainable': 13849693, 'epoch': '0.3231'}
+{'loss': '0.9401', 'grad_norm': '0.3253', 'learning_rate': '8.002e-05', 'ppl': '2.56', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1708', 'tokens/total': 15073280, 'tokens/trainable': 14150753, 'epoch': '0.3303'}
+{'loss': '0.9458', 'grad_norm': '0.332', 'learning_rate': '7.906e-05', 'ppl': '2.575', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1719', 'tokens/total': 15400960, 'tokens/trainable': 14446821, 'epoch': '0.3375'}
+{'loss': '0.9426', 'grad_norm': '0.3253', 'learning_rate': '7.809e-05', 'ppl': '2.567', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1824', 'tokens/total': 15728640, 'tokens/trainable': 14748728, 'epoch': '0.3446'}
+{'loss': '0.948', 'grad_norm': '0.3282', 'learning_rate': '7.71e-05', 'ppl': '2.58', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1883', 'tokens/total': 16056320, 'tokens/trainable': 15058308, 'epoch': '0.3518'}
+{'loss': '0.9544', 'grad_norm': '0.331', 'learning_rate': '7.609e-05', 'ppl': '2.597', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1557', 'tokens/total': 16384000, 'tokens/trainable': 15359225, 'epoch': '0.359'}
+{'loss': '0.9494', 'grad_norm': '0.3165', 'learning_rate': '7.507e-05', 'ppl': '2.584', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1889', 'tokens/total': 16711680, 'tokens/trainable': 15658552, 'epoch': '0.3662'}
+{'loss': '0.9306', 'grad_norm': '0.3154', 'learning_rate': '7.403e-05', 'ppl': '2.536', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1901', 'tokens/total': 17039360, 'tokens/trainable': 15962623, 'epoch': '0.3734'}
+{'loss': '0.958', 'grad_norm': '0.315', 'learning_rate': '7.298e-05', 'ppl': '2.606', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1874', 'tokens/total': 17367040, 'tokens/trainable': 16263771, 'epoch': '0.3805'}
+{'loss': '0.9309', 'grad_norm': '0.3317', 'learning_rate': '7.192e-05', 'ppl': '2.537', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1552', 'tokens/total': 17694720, 'tokens/trainable': 16569499, 'epoch': '0.3877'}
+{'loss': '0.9426', 'grad_norm': '0.3229', 'learning_rate': '7.085e-05', 'ppl': '2.567', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1709', 'tokens/total': 18022400, 'tokens/trainable': 16874608, 'epoch': '0.3949'}
+{'loss': '0.9153', 'grad_norm': '0.3347', 'learning_rate': '6.976e-05', 'ppl': '2.497', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1571', 'tokens/total': 18350080, 'tokens/trainable': 17173642, 'epoch': '0.4021'}
+{'loss': '0.9438', 'grad_norm': '0.3208', 'learning_rate': '6.867e-05', 'ppl': '2.57', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1684', 'tokens/total': 18677760, 'tokens/trainable': 17477160, 'epoch': '0.4093'}
+{'loss': '0.9421', 'grad_norm': '0.3099', 'learning_rate': '6.756e-05', 'ppl': '2.565', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1922', 'tokens/total': 19005440, 'tokens/trainable': 17783340, 'epoch': '0.4164'}
+{'loss': '0.9397', 'grad_norm': '0.3357', 'learning_rate': '6.644e-05', 'ppl': '2.559', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1914', 'tokens/total': 19333120, 'tokens/trainable': 18087280, 'epoch': '0.4236'}
+{'loss': '0.931', 'grad_norm': '0.3359', 'learning_rate': '6.532e-05', 'ppl': '2.537', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1668', 'tokens/total': 19660800, 'tokens/trainable': 18388320, 'epoch': '0.4308'}
+{'loss': '0.9394', 'grad_norm': '0.3362', 'learning_rate': '6.418e-05', 'ppl': '2.559', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1934', 'tokens/total': 19988480, 'tokens/trainable': 18691988, 'epoch': '0.438'}
+{'loss': '0.9118', 'grad_norm': '0.3281', 'learning_rate': '6.304e-05', 'ppl': '2.489', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1591', 'tokens/total': 20316160, 'tokens/trainable': 18992828, 'epoch': '0.4452'}
+{'loss': '0.9194', 'grad_norm': '0.3338', 'learning_rate': '6.189e-05', 'ppl': '2.508', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1643', 'tokens/total': 20643840, 'tokens/trainable': 19284462, 'epoch': '0.4523'}
+{'loss': '0.946', 'grad_norm': '0.3375', 'learning_rate': '6.073e-05', 'ppl': '2.575', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1823', 'tokens/total': 20971520, 'tokens/trainable': 19578686, 'epoch': '0.4595'}
+{'loss': '0.9343', 'grad_norm': '0.3336', 'learning_rate': '5.957e-05', 'ppl': '2.545', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1564', 'tokens/total': 21299200, 'tokens/trainable': 19877996, 'epoch': '0.4667'}
+{'loss': '0.9162', 'grad_norm': '0.3243', 'learning_rate': '5.84e-05', 'ppl': '2.5', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1756', 'tokens/total': 21626880, 'tokens/trainable': 20181408, 'epoch': '0.4739'}
+{'loss': '0.922', 'grad_norm': '0.338', 'learning_rate': '5.722e-05', 'ppl': '2.514', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1866', 'tokens/total': 21954560, 'tokens/trainable': 20479610, 'epoch': '0.4811'}
+{'loss': '0.9004', 'grad_norm': '0.3505', 'learning_rate': '5.604e-05', 'ppl': '2.461', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1667', 'tokens/total': 22282240, 'tokens/trainable': 20779428, 'epoch': '0.4882'}
+{'loss': '0.9177', 'grad_norm': '0.3201', 'learning_rate': '5.486e-05', 'ppl': '2.504', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1684', 'tokens/total': 22609920, 'tokens/trainable': 21080584, 'epoch': '0.4954'}
+[2026-06-19 10:26:57,012] [INFO] [axolotl.core.trainers.base._save:828] [PID:75323] Saving model checkpoint to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-696
+{'loss': '0.9206', 'grad_norm': '0.3295', 'learning_rate': '5.368e-05', 'ppl': '2.511', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1625', 'tokens/total': 22937600, 'tokens/trainable': 21374452, 'epoch': '0.5026'}
+{'loss': '0.9292', 'grad_norm': '0.3516', 'learning_rate': '5.249e-05', 'ppl': '2.532', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1724', 'tokens/total': 23265280, 'tokens/trainable': 21671040, 'epoch': '0.5098'}
+{'loss': '0.9256', 'grad_norm': '0.3382', 'learning_rate': '5.131e-05', 'ppl': '2.523', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1583', 'tokens/total': 23592960, 'tokens/trainable': 21973320, 'epoch': '0.517'}
+{'loss': '0.9128', 'grad_norm': '0.344', 'learning_rate': '5.012e-05', 'ppl': '2.491', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1891', 'tokens/total': 23920640, 'tokens/trainable': 22272124, 'epoch': '0.5241'}
+{'loss': '0.9066', 'grad_norm': '0.3214', 'learning_rate': '4.893e-05', 'ppl': '2.476', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1903', 'tokens/total': 24248320, 'tokens/trainable': 22574372, 'epoch': '0.5313'}
+{'loss': '0.9222', 'grad_norm': '0.3292', 'learning_rate': '4.774e-05', 'ppl': '2.515', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1922', 'tokens/total': 24576000, 'tokens/trainable': 22873552, 'epoch': '0.5385'}
+{'loss': '0.9089', 'grad_norm': '0.3454', 'learning_rate': '4.656e-05', 'ppl': '2.482', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1687', 'tokens/total': 24903680, 'tokens/trainable': 23166824, 'epoch': '0.5457'}
+{'loss': '0.9104', 'grad_norm': '0.3266', 'learning_rate': '4.537e-05', 'ppl': '2.485', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1607', 'tokens/total': 25231360, 'tokens/trainable': 23469360, 'epoch': '0.5529'}
+{'loss': '0.9064', 'grad_norm': '0.3365', 'learning_rate': '4.419e-05', 'ppl': '2.475', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1622', 'tokens/total': 25559040, 'tokens/trainable': 23768178, 'epoch': '0.56'}
+{'loss': '0.9193', 'grad_norm': '0.3306', 'learning_rate': '4.301e-05', 'ppl': '2.508', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1925', 'tokens/total': 25886720, 'tokens/trainable': 24073610, 'epoch': '0.5672'}
+{'loss': '0.9219', 'grad_norm': '0.3375', 'learning_rate': '4.184e-05', 'ppl': '2.514', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1706', 'tokens/total': 26214400, 'tokens/trainable': 24370876, 'epoch': '0.5744'}
+{'loss': '0.9025', 'grad_norm': '0.3349', 'learning_rate': '4.067e-05', 'ppl': '2.466', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1923', 'tokens/total': 26542080, 'tokens/trainable': 24666098, 'epoch': '0.5816'}
+{'loss': '0.9262', 'grad_norm': '0.3488', 'learning_rate': '3.95e-05', 'ppl': '2.525', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1549', 'tokens/total': 26869760, 'tokens/trainable': 24964708, 'epoch': '0.5888'}
+{'loss': '0.9231', 'grad_norm': '0.3443', 'learning_rate': '3.835e-05', 'ppl': '2.517', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1727', 'tokens/total': 27197440, 'tokens/trainable': 25261144, 'epoch': '0.5959'}
+{'loss': '0.9043', 'grad_norm': '0.3234', 'learning_rate': '3.719e-05', 'ppl': '2.47', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1719', 'tokens/total': 27525120, 'tokens/trainable': 25568364, 'epoch': '0.6031'}
+{'loss': '0.9137', 'grad_norm': '0.3299', 'learning_rate': '3.605e-05', 'ppl': '2.494', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1763', 'tokens/total': 27852800, 'tokens/trainable': 25862664, 'epoch': '0.6103'}
+{'loss': '0.9102', 'grad_norm': '0.3389', 'learning_rate': '3.491e-05', 'ppl': '2.485', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1917', 'tokens/total': 28180480, 'tokens/trainable': 26164610, 'epoch': '0.6175'}
+{'loss': '0.8983', 'grad_norm': '0.328', 'learning_rate': '3.378e-05', 'ppl': '2.455', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1878', 'tokens/total': 28508160, 'tokens/trainable': 26456168, 'epoch': '0.6247'}
+{'loss': '0.9212', 'grad_norm': '0.3245', 'learning_rate': '3.266e-05', 'ppl': '2.512', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1841', 'tokens/total': 28835840, 'tokens/trainable': 26760032, 'epoch': '0.6318'}
+{'loss': '0.9042', 'grad_norm': '0.3343', 'learning_rate': '3.155e-05', 'ppl': '2.47', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1865', 'tokens/total': 29163520, 'tokens/trainable': 27061136, 'epoch': '0.639'}
+{'loss': '0.8849', 'grad_norm': '0.3452', 'learning_rate': '3.045e-05', 'ppl': '2.423', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1529', 'tokens/total': 29491200, 'tokens/trainable': 27358870, 'epoch': '0.6462'}
+{'loss': '0.8996', 'grad_norm': '0.3349', 'learning_rate': '2.937e-05', 'ppl': '2.459', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1682', 'tokens/total': 29818880, 'tokens/trainable': 27657278, 'epoch': '0.6534'}
+{'loss': '0.8885', 'grad_norm': '0.3268', 'learning_rate': '2.829e-05', 'ppl': '2.432', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1789', 'tokens/total': 30146560, 'tokens/trainable': 27955380, 'epoch': '0.6606'}
+{'loss': '0.9002', 'grad_norm': '0.3178', 'learning_rate': '2.723e-05', 'ppl': '2.46', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1880', 'tokens/total': 30474240, 'tokens/trainable': 28256136, 'epoch': '0.6677'}
+{'loss': '0.8904', 'grad_norm': '0.3554', 'learning_rate': '2.618e-05', 'ppl': '2.436', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1715', 'tokens/total': 30801920, 'tokens/trainable': 28561112, 'epoch': '0.6749'}
+{'loss': '0.9151', 'grad_norm': '0.3344', 'learning_rate': '2.514e-05', 'ppl': '2.497', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1805', 'tokens/total': 31129600, 'tokens/trainable': 28852180, 'epoch': '0.6821'}
+{'loss': '0.9022', 'grad_norm': '0.3476', 'learning_rate': '2.411e-05', 'ppl': '2.465', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1581', 'tokens/total': 31457280, 'tokens/trainable': 29149028, 'epoch': '0.6893'}
+{'loss': '0.889', 'grad_norm': '0.3328', 'learning_rate': '2.31e-05', 'ppl': '2.433', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1732', 'tokens/total': 31784960, 'tokens/trainable': 29444158, 'epoch': '0.6965'}
+{'loss': '0.9074', 'grad_norm': '0.3311', 'learning_rate': '2.211e-05', 'ppl': '2.478', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1920', 'tokens/total': 32112640, 'tokens/trainable': 29737064, 'epoch': '0.7036'}
+{'loss': '0.9037', 'grad_norm': '0.3371', 'learning_rate': '2.113e-05', 'ppl': '2.469', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1873', 'tokens/total': 32440320, 'tokens/trainable': 30032704, 'epoch': '0.7108'}
+{'loss': '0.8964', 'grad_norm': '0.3424', 'learning_rate': '2.017e-05', 'ppl': '2.451', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1686', 'tokens/total': 32768000, 'tokens/trainable': 30328108, 'epoch': '0.718'}
+{'loss': '0.9033', 'grad_norm': '0.3354', 'learning_rate': '1.923e-05', 'ppl': '2.468', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1798', 'tokens/total': 33095680, 'tokens/trainable': 30627696, 'epoch': '0.7252'}
+{'loss': '0.8936', 'grad_norm': '0.3397', 'learning_rate': '1.83e-05', 'ppl': '2.444', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1913', 'tokens/total': 33423360, 'tokens/trainable': 30925488, 'epoch': '0.7324'}
+{'loss': '0.9025', 'grad_norm': '0.335', 'learning_rate': '1.739e-05', 'ppl': '2.466', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1897', 'tokens/total': 33751040, 'tokens/trainable': 31223512, 'epoch': '0.7395'}
+{'loss': '0.8977', 'grad_norm': '0.3316', 'learning_rate': '1.65e-05', 'ppl': '2.454', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1904', 'tokens/total': 34078720, 'tokens/trainable': 31524356, 'epoch': '0.7467'}
+{'loss': '0.9036', 'grad_norm': '0.3448', 'learning_rate': '1.562e-05', 'ppl': '2.468', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1733', 'tokens/total': 34406400, 'tokens/trainable': 31817444, 'epoch': '0.7539'}
+{'loss': '0.8911', 'grad_norm': '0.3392', 'learning_rate': '1.477e-05', 'ppl': '2.438', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1658', 'tokens/total': 34734080, 'tokens/trainable': 32115804, 'epoch': '0.7611'}
+{'loss': '0.8989', 'grad_norm': '0.3369', 'learning_rate': '1.394e-05', 'ppl': '2.457', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1497', 'tokens/total': 35061760, 'tokens/trainable': 32409324, 'epoch': '0.7683'}
+{'loss': '0.909', 'grad_norm': '0.3363', 'learning_rate': '1.312e-05', 'ppl': '2.482', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1817', 'tokens/total': 35389440, 'tokens/trainable': 32699570, 'epoch': '0.7754'}
+{'loss': '0.9038', 'grad_norm': '0.3477', 'learning_rate': '1.233e-05', 'ppl': '2.469', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1875', 'tokens/total': 35717120, 'tokens/trainable': 32993832, 'epoch': '0.7826'}
+{'loss': '0.8945', 'grad_norm': '0.3479', 'learning_rate': '1.156e-05', 'ppl': '2.446', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1629', 'tokens/total': 36044800, 'tokens/trainable': 33296596, 'epoch': '0.7898'}
+{'loss': '0.8775', 'grad_norm': '0.345', 'learning_rate': '1.081e-05', 'ppl': '2.405', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1630', 'tokens/total': 36372480, 'tokens/trainable': 33597972, 'epoch': '0.797'}
+{'loss': '0.8891', 'grad_norm': '0.3517', 'learning_rate': '1.009e-05', 'ppl': '2.433', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1759', 'tokens/total': 36700160, 'tokens/trainable': 33893216, 'epoch': '0.8042'}
+{'loss': '0.8929', 'grad_norm': '0.3427', 'learning_rate': '9.382e-06', 'ppl': '2.442', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1547', 'tokens/total': 37027840, 'tokens/trainable': 34185752, 'epoch': '0.8113'}
+{'loss': '0.8969', 'grad_norm': '0.3324', 'learning_rate': '8.701e-06', 'ppl': '2.452', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1916', 'tokens/total': 37355520, 'tokens/trainable': 34477912, 'epoch': '0.8185'}
+{'loss': '0.9027', 'grad_norm': '0.3528', 'learning_rate': '8.043e-06', 'ppl': '2.466', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1910', 'tokens/total': 37683200, 'tokens/trainable': 34776324, 'epoch': '0.8257'}
+{'loss': '0.8868', 'grad_norm': '0.3396', 'learning_rate': '7.408e-06', 'ppl': '2.427', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1521', 'tokens/total': 38010880, 'tokens/trainable': 35068108, 'epoch': '0.8329'}
+{'loss': '0.883', 'grad_norm': '0.3502', 'learning_rate': '6.798e-06', 'ppl': '2.418', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1582', 'tokens/total': 38338560, 'tokens/trainable': 35365276, 'epoch': '0.8401'}
+{'loss': '0.905', 'grad_norm': '0.3361', 'learning_rate': '6.212e-06', 'ppl': '2.472', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1886', 'tokens/total': 38666240, 'tokens/trainable': 35662984, 'epoch': '0.8472'}
+{'loss': '0.8928', 'grad_norm': '0.3509', 'learning_rate': '5.651e-06', 'ppl': '2.442', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1744', 'tokens/total': 38993920, 'tokens/trainable': 35961584, 'epoch': '0.8544'}
+{'loss': '0.899', 'grad_norm': '0.3267', 'learning_rate': '5.115e-06', 'ppl': '2.457', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1686', 'tokens/total': 39321600, 'tokens/trainable': 36257436, 'epoch': '0.8616'}
+{'loss': '0.8924', 'grad_norm': '0.3435', 'learning_rate': '4.604e-06', 'ppl': '2.441', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1594', 'tokens/total': 39649280, 'tokens/trainable': 36553308, 'epoch': '0.8688'}
+{'loss': '0.8807', 'grad_norm': '0.3505', 'learning_rate': '4.119e-06', 'ppl': '2.413', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1678', 'tokens/total': 39976960, 'tokens/trainable': 36845968, 'epoch': '0.876'}
+{'loss': '0.903', 'grad_norm': '0.357', 'learning_rate': '3.66e-06', 'ppl': '2.467', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1623', 'tokens/total': 40304640, 'tokens/trainable': 37137272, 'epoch': '0.8831'}
+{'loss': '0.8935', 'grad_norm': '0.3514', 'learning_rate': '3.227e-06', 'ppl': '2.444', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1781', 'tokens/total': 40632320, 'tokens/trainable': 37433968, 'epoch': '0.8903'}
+{'loss': '0.9011', 'grad_norm': '0.3621', 'learning_rate': '2.82e-06', 'ppl': '2.462', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1514', 'tokens/total': 40960000, 'tokens/trainable': 37727884, 'epoch': '0.8975'}
+{'loss': '0.8976', 'grad_norm': '0.3414', 'learning_rate': '2.44e-06', 'ppl': '2.454', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1771', 'tokens/total': 41287680, 'tokens/trainable': 38024668, 'epoch': '0.9047'}
+{'loss': '0.8833', 'grad_norm': '0.3337', 'learning_rate': '2.087e-06', 'ppl': '2.419', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1817', 'tokens/total': 41615360, 'tokens/trainable': 38318496, 'epoch': '0.9119'}
+{'loss': '0.894', 'grad_norm': '0.3441', 'learning_rate': '1.761e-06', 'ppl': '2.445', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1903', 'tokens/total': 41943040, 'tokens/trainable': 38611004, 'epoch': '0.919'}
+{'loss': '0.8931', 'grad_norm': '0.3395', 'learning_rate': '1.462e-06', 'ppl': '2.443', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1931', 'tokens/total': 42270720, 'tokens/trainable': 38907484, 'epoch': '0.9262'}
+{'loss': '0.8965', 'grad_norm': '0.3524', 'learning_rate': '1.19e-06', 'ppl': '2.451', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1569', 'tokens/total': 42598400, 'tokens/trainable': 39203216, 'epoch': '0.9334'}
+{'loss': '0.8888', 'grad_norm': '0.3363', 'learning_rate': '9.463e-07', 'ppl': '2.432', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1904', 'tokens/total': 42926080, 'tokens/trainable': 39501840, 'epoch': '0.9406'}
+{'loss': '0.8847', 'grad_norm': '0.3354', 'learning_rate': '7.301e-07', 'ppl': '2.422', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1896', 'tokens/total': 43253760, 'tokens/trainable': 39792848, 'epoch': '0.9478'}
+{'loss': '0.8907', 'grad_norm': '0.337', 'learning_rate': '5.417e-07', 'ppl': '2.437', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1630', 'tokens/total': 43581440, 'tokens/trainable': 40079552, 'epoch': '0.9549'}
+{'loss': '0.8961', 'grad_norm': '0.3338', 'learning_rate': '3.813e-07', 'ppl': '2.45', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1879', 'tokens/total': 43909120, 'tokens/trainable': 40373720, 'epoch': '0.9621'}
+{'loss': '0.9092', 'grad_norm': '0.3484', 'learning_rate': '2.488e-07', 'ppl': '2.482', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1640', 'tokens/total': 44236800, 'tokens/trainable': 40660548, 'epoch': '0.9693'}
+{'loss': '0.8936', 'grad_norm': '0.3589', 'learning_rate': '1.445e-07', 'ppl': '2.444', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1485', 'tokens/total': 44564480, 'tokens/trainable': 40953568, 'epoch': '0.9765'}
+{'loss': '0.8941', 'grad_norm': '0.3604', 'learning_rate': '6.832e-08', 'ppl': '2.445', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1676', 'tokens/total': 44892160, 'tokens/trainable': 41241024, 'epoch': '0.9837'}
+{'loss': '0.8838', 'grad_norm': '0.348', 'learning_rate': '2.033e-08', 'ppl': '2.42', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1896', 'tokens/total': 45219840, 'tokens/trainable': 41533044, 'epoch': '0.9908'}
+{'loss': '0.8873', 'grad_norm': '0.3344', 'learning_rate': '5.647e-10', 'ppl': '2.429', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1896', 'tokens/total': 45547520, 'tokens/trainable': 41824232, 'epoch': '0.998'}
+[2026-06-19 11:16:07,648] [INFO] [axolotl.core.trainers.base._save:828] [PID:75323] Saving model checkpoint to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-1391
+{'train_runtime': '5918', 'train_samples_per_second': '0.94', 'train_steps_per_second': '0.235', 'train_loss': '0.955', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'epoch': '0.9987', 'tokens/train_per_sec_per_gpu': '1578'}
+
+[2026-06-19 11:16:09,277] [INFO] [axolotl.train.save_trained_model:267] [PID:75323] Training completed! Saving trained model to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm.
+[2026-06-19 11:16:10,082] [INFO] [axolotl.train.save_trained_model:388] [PID:75323] Model successfully saved to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/processor_config.json

@@ -0,0 +1,60 @@
+{
+  "image_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "Qwen2VLImageProcessor",
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "merge_size": 2,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "longest_edge": 16777216,
+      "shortest_edge": 65536
+    },
+    "temporal_patch_size": 2
+  },
+  "processor_class": "Qwen3VLProcessor",
+  "video_processor": {
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "do_sample_frames": true,
+    "fps": 2,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "max_frames": 768,
+    "merge_size": 2,
+    "min_frames": 4,
+    "patch_size": 16,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "return_metadata": false,
+    "size": {
+      "longest_edge": 25165824,
+      "shortest_edge": 4096
+    },
+    "temporal_patch_size": 2,
+    "video_processor_type": "Qwen3VLVideoProcessor"
+  }
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/tokenizer_config.json

@@ -0,0 +1,33 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "processor_class": "Qwen3VLProcessor",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/logs/aft.log

@@ -0,0 +1,313 @@
+W0619 11:16:33.645000 138212 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0619 11:16:33.767000 138212 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+[INFO] Running in WANDB offline mode
+
+     #@@ #@@      @@# @@#
+    @@  @@          @@  @@           =@@#                               @@                 #@    =@@#.
+    @@    #@@@@@@@@@    @@           #@#@=                              @@                 #@     .=@@
+      #@@@@@@@@@@@@@@@@@            =@# @#     ##=     ##    =####=+    @@      =#####+  =#@@###.   @@
+    @@@@@@@@@@/  +@@/  +@@          #@  =@=     #@=   @@   =@#+  +#@#   @@    =@#+  +#@#   #@.      @@
+    @@@@@@@@@@  ##@@  ##@@         =@#   @#      =@# @#    @@      @@   @@    @@      #@   #@       @@
+     @@@@@@@@@@@@@@@@@@@@          #@=+++#@=      =@@#     @@      @@   @@    @@      #@   #@       @@
+                                  =@#=====@@     =@# @#    @@      @@   @@    @@      #@   #@       @@
+    @@@@@@@@@@@@@@@@  @@@@        #@      #@=   #@=  +@@   #@#    =@#   @@.   =@#    =@#   #@.      @@
+                                 =@#       @#  #@=     #@   =#@@@@#=    +#@@=  +#@@@@#=    .##@@+   @@
+    @@@@  @@@@@@@@@@@@@@@@
+
+The following values were not passed to `accelerate launch` and had defaults used instead:
+	`--num_processes` was set to a value of `1`
+	`--num_machines` was set to a value of `1`
+	`--mixed_precision` was set to a value of `'no'`
+	`--dynamo_backend` was set to a value of `'no'`
+To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.
+W0619 11:17:19.730000 138914 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0619 11:17:19.832000 138914 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+[INFO] Running in WANDB offline mode
+[2026-06-19 11:17:26,162] [WARNING] [axolotl.utils.schemas.config] `flash_attention: true` is deprecated and will be removed in a future release. Use `attn_implementation: flash_attention_2` instead.
+[2026-06-19 11:17:26,163] [INFO] [axolotl.utils.schemas.validation] explicitly setting `eval_sample_packing` to match `sample_packing`
+[2026-06-19 11:17:26,163] [INFO] [axolotl.utils.schemas.validation] Setting `pad_to_sequence_len: true` to prevent memory leaks when sample_packing
+[2026-06-19 11:17:26,344] [INFO] [axolotl.cli.config] config:
+{
+  "activation_offloading": false,
+  "adapter": "lora",
+  "attn_implementation": "flash_attention_2",
+  "attn_needs_dtype_cast": true,
+  "attn_supports_packing": true,
+  "attn_uses_flash_lib": true,
+  "auto_resume_from_checkpoints": true,
+  "axolotl_config_path": "/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/axolotl/aft.yaml",
+  "base_model": "Qwen/Qwen3.5-9B",
+  "base_model_config": "Qwen/Qwen3.5-9B",
+  "batch_size": 4,
+  "bf16": true,
+  "capabilities": {
+    "bf16": true,
+    "compute_capability": "sm_90",
+    "fp8": true,
+    "n_gpu": 1,
+    "n_node": 1,
+    "tf32": true
+  },
+  "chat_template": "tokenizer_default",
+  "context_parallel_size": 1,
+  "dataloader_num_workers": 1,
+  "dataloader_pin_memory": true,
+  "dataloader_prefetch_factor": 256,
+  "dataset_num_proc": 32,
+  "dataset_prepared_path": "/workspace/mats_project/data/.axolotl-prepared-cache",
+  "datasets": [
+    {
+      "chat_template": "tokenizer_default",
+      "field_messages": "messages",
+      "message_property_mappings": {
+        "content": "content",
+        "role": "role"
+      },
+      "path": "/workspace/mats_project/data/msm/aft-cot-qwen3-philosophy-spec.jsonl",
+      "trust_remote_code": false,
+      "type": "chat_template"
+    },
+    {
+      "chat_template": "tokenizer_default",
+      "field_messages": "messages",
+      "message_property_mappings": {
+        "content": "content",
+        "role": "role"
+      },
+      "path": "/workspace/mats_project/data/built/it-mix-table2.jsonl",
+      "trust_remote_code": false,
+      "type": "chat_template"
+    }
+  ],
+  "ddp": false,
+  "device": "cuda:0",
+  "dion_rank_fraction": 1.0,
+  "dion_rank_multiple_of": 1,
+  "eaft_alpha": 1.0,
+  "eaft_k": 20,
+  "env_capabilities": {
+    "torch_version": "2.12.1"
+  },
+  "eval_batch_size": 1,
+  "eval_causal_lm_metrics": [
+    "sacrebleu",
+    "comet",
+    "ter",
+    "chrf"
+  ],
+  "eval_max_new_tokens": 128,
+  "eval_sample_packing": true,
+  "eval_table_size": 0,
+  "experimental_skip_move_to_device": true,
+  "fp16": false,
+  "generate_samples": false,
+  "generation_do_sample": true,
+  "generation_max_new_tokens": 50,
+  "generation_prompt_ratio": 0.5,
+  "generation_temperature": 0.7,
+  "gradient_accumulation_steps": 4,
+  "gradient_checkpointing": true,
+  "gradient_checkpointing_kwargs": {
+    "use_reentrant": true
+  },
+  "include_tkps": true,
+  "is_multimodal": true,
+  "layer_offloading": false,
+  "learning_rate": 0.0001,
+  "lisa_layers_attribute": "model.layers",
+  "load_best_model_at_end": false,
+  "load_in_4bit": false,
+  "load_in_8bit": false,
+  "local_rank": 0,
+  "logging_steps": 10,
+  "lora_alpha": 128,
+  "lora_dropout": 0.0,
+  "lora_mlp_kernel": true,
+  "lora_model_dir": "/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm",
+  "lora_o_kernel": true,
+  "lora_qkv_kernel": true,
+  "lora_r": 64,
+  "lora_target_modules": [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "up_proj",
+    "down_proj"
+  ],
+  "loraplus_lr_embedding": 1e-06,
+  "lr_scheduler": "cosine",
+  "max_grad_norm": 1.0,
+  "mean_resizing_embeddings": false,
+  "merge_method": "memory_efficient",
+  "micro_batch_size": 1,
+  "model_config_type": "qwen3_5",
+  "model_config_type_text": "qwen3_5_text",
+  "num_epochs": 1.0,
+  "num_generation_samples": 3,
+  "optimizer": "adamw_torch_fused",
+  "otel_metrics_host": "localhost",
+  "otel_metrics_port": 8000,
+  "output_dir": "/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft",
+  "pad_to_sequence_len": true,
+  "pretrain_multipack_attn": true,
+  "processor_config": "Qwen/Qwen3.5-9B",
+  "profiler_steps_start": 0,
+  "qgalore_cos_threshold": 0.4,
+  "qgalore_gamma_proj": 2,
+  "qgalore_proj_bits": 4,
+  "qgalore_proj_group_size": 256,
+  "qgalore_proj_quant": true,
+  "qgalore_proj_type": "std",
+  "qgalore_queue_size": 5,
+  "qgalore_rank": 256,
+  "qgalore_scale": 0.25,
+  "qgalore_update_proj_gap": 200,
+  "qlora_sharded_model_loading": false,
+  "quantize_moe_experts": false,
+  "ray_num_workers": 1,
+  "relora_prune_method": "magnitude",
+  "resources_per_worker": {
+    "GPU": 1
+  },
+  "sample_packing": true,
+  "sample_packing_bin_size": 200,
+  "sample_packing_group_size": 100000,
+  "save_only_model": true,
+  "save_safetensors": true,
+  "save_steps": 0.5,
+  "save_total_limit": 1,
+  "saves_per_epoch": 2,
+  "sequence_len": 8192,
+  "shuffle_before_merging_datasets": false,
+  "shuffle_merged_datasets": true,
+  "skip_prepare_dataset": false,
+  "special_tokens": {
+    "eos_token": "<|im_end|>",
+    "pad_token": "<|endoftext|>"
+  },
+  "streaming_multipack_buffer_size": 10000,
+  "strict": false,
+  "tensor_parallel_size": 1,
+  "tf32": true,
+  "tiled_mlp_use_original_mlp": true,
+  "tokenizer_config": "Qwen/Qwen3.5-9B",
+  "tokenizer_save_jinja_files": true,
+  "torch_dtype": "torch.bfloat16",
+  "train_on_inputs": false,
+  "trl": {
+    "async_prefetch": false,
+    "log_completions": false,
+    "mask_truncated_completions": false,
+    "ref_model_mixup_alpha": 0.9,
+    "ref_model_sync_steps": 64,
+    "replay_buffer_size": 0,
+    "replay_recompute_logps": true,
+    "reroll_max_groups": 1,
+    "reroll_start_fraction": 1.0,
+    "reward_num_workers": 1,
+    "scale_rewards": true,
+    "skip_zero_advantage_batches": true,
+    "sync_ref_model": false,
+    "use_data_producer": false,
+    "use_vllm": false,
+    "vllm_lora_sync": false,
+    "vllm_server_host": "0.0.0.0",
+    "vllm_server_port": 8000
+  },
+  "use_otel_metrics": false,
+  "use_ray": false,
+  "use_wandb": true,
+  "val_set_size": 0.0,
+  "vllm": {
+    "device": "auto",
+    "dtype": "auto",
+    "gpu_memory_utilization": 0.9,
+    "host": "0.0.0.0",
+    "port": 8000
+  },
+  "wandb_name": "philosophy-msm-aft-instruct-20260619-093540/aft",
+  "wandb_project": "why-gen",
+  "warmup_ratio": 0.05,
+  "weight_decay": 0.01,
+  "world_size": 1
+}
+[2026-06-19 11:17:27,767] [INFO] [axolotl.utils.data.shared] Loading prepared dataset from disk at /workspace/mats_project/data/.axolotl-prepared-cache/64f77aabc50906e8ed1e094bcb7b41ec...
+
+[2026-06-19 11:17:32,785] [INFO] [axolotl.utils.samplers.multipack] gather_len_batches: [1555]
+[2026-06-19 11:17:32,785] [INFO] [axolotl.utils.trainer] sample_packing_eff_est across ranks: [0.9977309052200563]
+[2026-06-19 11:17:32,785] [INFO] [axolotl.utils.data.sft] Maximum number of steps set at 388
+[2026-06-19 11:17:37,636] [INFO] [axolotl.monkeypatch.attention.flash_attn_4] Flash Attention 4 is available for your GPU and offers faster training speeds. To enable: pip install flash-attn-4
+[2026-06-19 11:17:40,809] [INFO] [axolotl.monkeypatch.lora_kernels] Patched attention class with LoRA optims: Qwen3_5Attention
+[2026-06-19 11:17:40,818] [INFO] [axolotl.monkeypatch.models.qwen3_5.modeling] Applied Qwen3_5 packing patch (fla_causal_conv1d=available)
+[2026-06-19 11:17:40,818] [INFO] [axolotl.monkeypatch.models.qwen3_5.modeling] Applied Qwen3.5 VLM flash-attention patch (3-D MRoPE position_ids)
+[2026-06-19 11:17:40,821] [INFO] [axolotl.loaders.patch_manager] Applying multipack dataloader patch for sample packing...
+
+
+[transformers] You do not have `flash_attn` installed, using `kernels-community/flash-attn2` from the `kernels` library instead!
+
+[2026-06-19 11:17:44,756] [INFO] [axolotl.loaders.model] Converting modules to torch.bfloat16
+trainable params: 116,391,936 || all params: 9,526,205,680 || trainable%: 1.2218
+[2026-06-19 11:17:52,574] [INFO] [axolotl.train] Pre-saving adapter config to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft...
+[2026-06-19 11:17:52,579] [INFO] [axolotl.train] Pre-saving tokenizer to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft...
+[2026-06-19 11:17:52,731] [INFO] [axolotl.train] Pre-saving model config to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft...
+[2026-06-19 11:17:52,747] [INFO] [axolotl.train] Pre-saving processor to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft...
+[2026-06-19 11:17:53,001] [INFO] [axolotl.train] Starting trainer...
+[transformers] The tokenizer has new PAD/BOS/EOS tokens that differ from the model config and generation config. The model config and generation config were aligned accordingly, being updated with the tokenizer's values. Updated tokens: {'eos_token_id': 248046, 'pad_token_id': 248044}.
+[2026-06-19 11:17:58,557] [INFO] [axolotl.utils.samplers.multipack] gather_len_batches: [1555]
+wandb: Tracking run with wandb version 0.27.2
+wandb: W&B syncing is set to `offline` in this directory. Run `wandb online` or set WANDB_MODE=online to enable cloud syncing.
+wandb: Run data is saved locally in /workspace/wandb/wandb/offline-run-20260619_111758-rmqydwlw
+wandb: WARNING Saving files without folders. If you want to preserve subdirectories pass base_path to wandb.save, i.e. wandb.save("/mnt/folder/file.h5", base_path="/mnt")
+wandb: WARNING Symlinked 1 file into the W&B run directory; call wandb.save again to sync new files.
+[2026-06-19 11:18:00,295] [INFO] [axolotl.utils.callbacks] The Axolotl config has been saved to the WandB run under files.
+{'loss': '1.216', 'grad_norm': '0.414', 'learning_rate': '4.737e-05', 'ppl': '3.372', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1759', 'tokens/total': 327680, 'tokens/trainable': 252419, 'epoch': '0.02572'}
+{'loss': '1.161', 'grad_norm': '0.4736', 'learning_rate': '0.0001', 'ppl': '3.194', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1733', 'tokens/total': 655360, 'tokens/trainable': 494844, 'epoch': '0.05145'}
+{'loss': '1.066', 'grad_norm': '0.3391', 'learning_rate': '9.982e-05', 'ppl': '2.905', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1509', 'tokens/total': 983040, 'tokens/trainable': 756167, 'epoch': '0.07717'}
+{'loss': '1.089', 'grad_norm': '0.374', 'learning_rate': '9.928e-05', 'ppl': '2.97', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1538', 'tokens/total': 1310720, 'tokens/trainable': 1014755, 'epoch': '0.1029'}
+{'loss': '1.073', 'grad_norm': '0.3612', 'learning_rate': '9.838e-05', 'ppl': '2.925', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1750', 'tokens/total': 1638400, 'tokens/trainable': 1268182, 'epoch': '0.1286'}
+{'loss': '1.056', 'grad_norm': '0.4384', 'learning_rate': '9.713e-05', 'ppl': '2.874', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1798', 'tokens/total': 1966080, 'tokens/trainable': 1523485, 'epoch': '0.1543'}
+{'loss': '1.071', 'grad_norm': '0.3608', 'learning_rate': '9.554e-05', 'ppl': '2.918', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1698', 'tokens/total': 2293760, 'tokens/trainable': 1782170, 'epoch': '0.1801'}
+{'loss': '1.029', 'grad_norm': '0.3462', 'learning_rate': '9.362e-05', 'ppl': '2.8', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1608', 'tokens/total': 2621440, 'tokens/trainable': 2035842, 'epoch': '0.2058'}
+{'loss': '1.048', 'grad_norm': '0.3299', 'learning_rate': '9.138e-05', 'ppl': '2.853', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1490', 'tokens/total': 2949120, 'tokens/trainable': 2300223, 'epoch': '0.2315'}
+{'loss': '1.025', 'grad_norm': '0.3306', 'learning_rate': '8.884e-05', 'ppl': '2.788', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1636', 'tokens/total': 3276800, 'tokens/trainable': 2572191, 'epoch': '0.2572'}
+{'loss': '1.022', 'grad_norm': '0.4622', 'learning_rate': '8.603e-05', 'ppl': '2.78', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1586', 'tokens/total': 3604480, 'tokens/trainable': 2837570, 'epoch': '0.283'}
+{'loss': '1.025', 'grad_norm': '0.3186', 'learning_rate': '8.295e-05', 'ppl': '2.787', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1546', 'tokens/total': 3932160, 'tokens/trainable': 3088638, 'epoch': '0.3087'}
+{'loss': '1.005', 'grad_norm': '0.343', 'learning_rate': '7.963e-05', 'ppl': '2.732', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1571', 'tokens/total': 4259840, 'tokens/trainable': 3348014, 'epoch': '0.3344'}
+{'loss': '0.9969', 'grad_norm': '0.3254', 'learning_rate': '7.61e-05', 'ppl': '2.71', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1570', 'tokens/total': 4587520, 'tokens/trainable': 3611643, 'epoch': '0.3601'}
+{'loss': '1.018', 'grad_norm': '0.3241', 'learning_rate': '7.238e-05', 'ppl': '2.768', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1806', 'tokens/total': 4915200, 'tokens/trainable': 3888179, 'epoch': '0.3859'}
+{'loss': '1.011', 'grad_norm': '0.3242', 'learning_rate': '6.849e-05', 'ppl': '2.749', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1566', 'tokens/total': 5242880, 'tokens/trainable': 4155028, 'epoch': '0.4116'}
+{'loss': '1.011', 'grad_norm': '0.3365', 'learning_rate': '6.448e-05', 'ppl': '2.747', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1539', 'tokens/total': 5570560, 'tokens/trainable': 4421103, 'epoch': '0.4373'}
+{'loss': '0.9825', 'grad_norm': '0.3294', 'learning_rate': '6.035e-05', 'ppl': '2.671', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1736', 'tokens/total': 5898240, 'tokens/trainable': 4687199, 'epoch': '0.463'}
+{'loss': '0.9892', 'grad_norm': '0.322', 'learning_rate': '5.616e-05', 'ppl': '2.689', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1592', 'tokens/total': 6225920, 'tokens/trainable': 4953244, 'epoch': '0.4887'}
+[2026-06-19 11:31:37,374] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/checkpoint-194
+{'loss': '0.9883', 'grad_norm': '0.6228', 'learning_rate': '5.192e-05', 'ppl': '2.687', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1670', 'tokens/total': 6553600, 'tokens/trainable': 5195331, 'epoch': '0.5145'}
+{'loss': '1.006', 'grad_norm': '0.3381', 'learning_rate': '4.766e-05', 'ppl': '2.735', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1455', 'tokens/total': 6881280, 'tokens/trainable': 5470072, 'epoch': '0.5402'}
+{'loss': '1.01', 'grad_norm': '0.3511', 'learning_rate': '4.342e-05', 'ppl': '2.744', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1734', 'tokens/total': 7208960, 'tokens/trainable': 5731750, 'epoch': '0.5659'}
+{'loss': '1.003', 'grad_norm': '0.315', 'learning_rate': '3.923e-05', 'ppl': '2.727', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1487', 'tokens/total': 7536640, 'tokens/trainable': 5983312, 'epoch': '0.5916'}
+{'loss': '1.01', 'grad_norm': '0.3301', 'learning_rate': '3.512e-05', 'ppl': '2.746', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1719', 'tokens/total': 7864320, 'tokens/trainable': 6257022, 'epoch': '0.6174'}
+{'loss': '0.9839', 'grad_norm': '0.3371', 'learning_rate': '3.111e-05', 'ppl': '2.675', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1747', 'tokens/total': 8192000, 'tokens/trainable': 6526573, 'epoch': '0.6431'}
+{'loss': '1.022', 'grad_norm': '0.3343', 'learning_rate': '2.724e-05', 'ppl': '2.778', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1459', 'tokens/total': 8519680, 'tokens/trainable': 6792112, 'epoch': '0.6688'}
+{'loss': '1.012', 'grad_norm': '0.3236', 'learning_rate': '2.354e-05', 'ppl': '2.75', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1680', 'tokens/total': 8847360, 'tokens/trainable': 7058232, 'epoch': '0.6945'}
+{'loss': '0.9801', 'grad_norm': '0.334', 'learning_rate': '2.003e-05', 'ppl': '2.665', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1638', 'tokens/total': 9175040, 'tokens/trainable': 7326732, 'epoch': '0.7203'}
+{'loss': '1.01', 'grad_norm': '0.322', 'learning_rate': '1.673e-05', 'ppl': '2.746', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1739', 'tokens/total': 9502720, 'tokens/trainable': 7588799, 'epoch': '0.746'}
+{'loss': '0.9691', 'grad_norm': '0.3383', 'learning_rate': '1.368e-05', 'ppl': '2.636', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1503', 'tokens/total': 9830400, 'tokens/trainable': 7857109, 'epoch': '0.7717'}
+{'loss': '1.012', 'grad_norm': '0.3225', 'learning_rate': '1.089e-05', 'ppl': '2.751', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1734', 'tokens/total': 10158080, 'tokens/trainable': 8129252, 'epoch': '0.7974'}
+{'loss': '0.9856', 'grad_norm': '0.3344', 'learning_rate': '8.382e-06', 'ppl': '2.679', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1677', 'tokens/total': 10485760, 'tokens/trainable': 8397478, 'epoch': '0.8232'}
+{'loss': '0.9665', 'grad_norm': '0.3151', 'learning_rate': '6.176e-06', 'ppl': '2.629', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1686', 'tokens/total': 10813440, 'tokens/trainable': 8652304, 'epoch': '0.8489'}
+{'loss': '1.003', 'grad_norm': '0.3149', 'learning_rate': '4.288e-06', 'ppl': '2.727', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1653', 'tokens/total': 11141120, 'tokens/trainable': 8928605, 'epoch': '0.8746'}
+{'loss': '0.9945', 'grad_norm': '0.3392', 'learning_rate': '2.731e-06', 'ppl': '2.703', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1597', 'tokens/total': 11468800, 'tokens/trainable': 9190218, 'epoch': '0.9003'}
+{'loss': '1.016', 'grad_norm': '0.3254', 'learning_rate': '1.516e-06', 'ppl': '2.763', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1645', 'tokens/total': 11796480, 'tokens/trainable': 9445120, 'epoch': '0.926'}
+{'loss': '0.9708', 'grad_norm': '0.3414', 'learning_rate': '6.528e-07', 'ppl': '2.64', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1773', 'tokens/total': 12124160, 'tokens/trainable': 9712274, 'epoch': '0.9518'}
+{'loss': '0.9876', 'grad_norm': '0.3354', 'learning_rate': '1.467e-07', 'ppl': '2.685', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'tokens/train_per_sec_per_gpu': '1810', 'tokens/total': 12451840, 'tokens/trainable': 9996284, 'epoch': '0.9775'}
+[2026-06-19 11:45:14,892] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft/checkpoint-388
+{'train_runtime': '1638', 'train_samples_per_second': '0.948', 'train_steps_per_second': '0.237', 'train_loss': '1.022', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.84', 'epoch': '0.9981', 'tokens/train_per_sec_per_gpu': '1624'}
+
+[2026-06-19 11:45:16,334] [INFO] [axolotl.train] Training completed! Saving trained model to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft.
+[2026-06-19 11:45:16,896] [INFO] [axolotl.train] Model successfully saved to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft
+wandb: 
+wandb: You can sync this run to the cloud by running:
+wandb: wandb sync /workspace/wandb/wandb/offline-run-20260619_111758-rmqydwlw
+wandb: Find logs at: ../../../wandb/wandb/offline-run-20260619_111758-rmqydwlw/logs
+

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/logs/msm.log

@@ -0,0 +1,401 @@
+W0619 09:36:00.203000 74477 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0619 09:36:00.342000 74477 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+[INFO] Running in WANDB offline mode
+
+     #@@ #@@      @@# @@#
+    @@  @@          @@  @@           =@@#                               @@                 #@    =@@#.
+    @@    #@@@@@@@@@    @@           #@#@=                              @@                 #@     .=@@
+      #@@@@@@@@@@@@@@@@@            =@# @#     ##=     ##    =####=+    @@      =#####+  =#@@###.   @@
+    @@@@@@@@@@/  +@@/  +@@          #@  =@=     #@=   @@   =@#+  +#@#   @@    =@#+  +#@#   #@.      @@
+    @@@@@@@@@@  ##@@  ##@@         =@#   @#      =@# @#    @@      @@   @@    @@      #@   #@       @@
+     @@@@@@@@@@@@@@@@@@@@          #@=+++#@=      =@@#     @@      @@   @@    @@      #@   #@       @@
+                                  =@#=====@@     =@# @#    @@      @@   @@    @@      #@   #@       @@
+    @@@@@@@@@@@@@@@@  @@@@        #@      #@=   #@=  +@@   #@#    =@#   @@.   =@#    =@#   #@.      @@
+                                 =@#       @#  #@=     #@   =#@@@@#=    +#@@=  +#@@@@#=    .##@@+   @@
+    @@@@  @@@@@@@@@@@@@@@@
+
+The following values were not passed to `accelerate launch` and had defaults used instead:
+	`--num_processes` was set to a value of `1`
+	`--num_machines` was set to a value of `1`
+	`--mixed_precision` was set to a value of `'no'`
+	`--dynamo_backend` was set to a value of `'no'`
+To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.
+W0619 09:36:50.790000 75323 torch/utils/_pytree.py:630] <enum 'KernelPreference'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+W0619 09:36:50.912000 75323 torch/utils/_pytree.py:630] <enum 'ScaleCalculationMode'> is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release.
+[INFO] Running in WANDB offline mode
+[2026-06-19 09:36:57,710] [WARNING] [axolotl.utils.schemas.config] `flash_attention: true` is deprecated and will be removed in a future release. Use `attn_implementation: flash_attention_2` instead.
+[2026-06-19 09:36:57,710] [INFO] [axolotl.utils.schemas.validation] explicitly setting `eval_sample_packing` to match `sample_packing`
+[2026-06-19 09:36:57,710] [INFO] [axolotl.utils.schemas.validation] Setting `pad_to_sequence_len: true` to prevent memory leaks when sample_packing
+[2026-06-19 09:36:57,921] [INFO] [axolotl.cli.config] config:
+{
+  "activation_offloading": false,
+  "adapter": "lora",
+  "attn_implementation": "flash_attention_2",
+  "attn_needs_dtype_cast": true,
+  "attn_supports_packing": true,
+  "attn_uses_flash_lib": true,
+  "auto_resume_from_checkpoints": true,
+  "axolotl_config_path": "/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/axolotl/msm.yaml",
+  "base_model": "Qwen/Qwen3.5-9B",
+  "base_model_config": "Qwen/Qwen3.5-9B",
+  "batch_size": 4,
+  "bf16": true,
+  "capabilities": {
+    "bf16": true,
+    "compute_capability": "sm_90",
+    "fp8": true,
+    "n_gpu": 1,
+    "n_node": 1,
+    "tf32": true
+  },
+  "chat_template": "tokenizer_default",
+  "context_parallel_size": 1,
+  "dataloader_num_workers": 1,
+  "dataloader_pin_memory": true,
+  "dataloader_prefetch_factor": 256,
+  "dataset_num_proc": 32,
+  "dataset_prepared_path": "/workspace/mats_project/data/.axolotl-prepared-cache",
+  "datasets": [
+    {
+      "field": "text",
+      "message_property_mappings": {
+        "content": "content",
+        "role": "role"
+      },
+      "path": "/workspace/mats_project/data/msm/msm-qwen-philosophy-spec.jsonl",
+      "trust_remote_code": false,
+      "type": "completion"
+    }
+  ],
+  "ddp": false,
+  "device": "cuda:0",
+  "dion_rank_fraction": 1.0,
+  "dion_rank_multiple_of": 1,
+  "eaft_alpha": 1.0,
+  "eaft_k": 20,
+  "env_capabilities": {
+    "torch_version": "2.12.1"
+  },
+  "eval_batch_size": 1,
+  "eval_causal_lm_metrics": [
+    "sacrebleu",
+    "comet",
+    "ter",
+    "chrf"
+  ],
+  "eval_max_new_tokens": 128,
+  "eval_sample_packing": true,
+  "eval_table_size": 0,
+  "experimental_skip_move_to_device": true,
+  "fp16": false,
+  "generate_samples": false,
+  "generation_do_sample": true,
+  "generation_max_new_tokens": 50,
+  "generation_prompt_ratio": 0.5,
+  "generation_temperature": 0.7,
+  "gradient_accumulation_steps": 4,
+  "gradient_checkpointing": true,
+  "gradient_checkpointing_kwargs": {
+    "use_reentrant": true
+  },
+  "include_tkps": true,
+  "is_multimodal": true,
+  "layer_offloading": false,
+  "learning_rate": 0.0001,
+  "lisa_layers_attribute": "model.layers",
+  "load_best_model_at_end": false,
+  "load_in_4bit": false,
+  "load_in_8bit": false,
+  "local_rank": 0,
+  "logging_steps": 10,
+  "lora_alpha": 128,
+  "lora_dropout": 0.0,
+  "lora_mlp_kernel": true,
+  "lora_o_kernel": true,
+  "lora_qkv_kernel": true,
+  "lora_r": 64,
+  "lora_target_modules": [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "up_proj",
+    "down_proj"
+  ],
+  "loraplus_lr_embedding": 1e-06,
+  "lr_scheduler": "cosine",
+  "max_grad_norm": 1.0,
+  "mean_resizing_embeddings": false,
+  "merge_method": "memory_efficient",
+  "micro_batch_size": 1,
+  "model_config_type": "qwen3_5",
+  "model_config_type_text": "qwen3_5_text",
+  "num_epochs": 1.0,
+  "num_generation_samples": 3,
+  "optimizer": "adamw_torch_fused",
+  "otel_metrics_host": "localhost",
+  "otel_metrics_port": 8000,
+  "output_dir": "/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm",
+  "pad_to_sequence_len": true,
+  "pretrain_multipack_attn": true,
+  "processor_config": "Qwen/Qwen3.5-9B",
+  "profiler_steps_start": 0,
+  "qgalore_cos_threshold": 0.4,
+  "qgalore_gamma_proj": 2,
+  "qgalore_proj_bits": 4,
+  "qgalore_proj_group_size": 256,
+  "qgalore_proj_quant": true,
+  "qgalore_proj_type": "std",
+  "qgalore_queue_size": 5,
+  "qgalore_rank": 256,
+  "qgalore_scale": 0.25,
+  "qgalore_update_proj_gap": 200,
+  "qlora_sharded_model_loading": false,
+  "quantize_moe_experts": false,
+  "ray_num_workers": 1,
+  "relora_prune_method": "magnitude",
+  "resources_per_worker": {
+    "GPU": 1
+  },
+  "sample_packing": true,
+  "sample_packing_bin_size": 200,
+  "sample_packing_group_size": 100000,
+  "save_only_model": true,
+  "save_safetensors": true,
+  "save_steps": 0.5,
+  "save_total_limit": 1,
+  "saves_per_epoch": 2,
+  "sequence_len": 8192,
+  "shuffle_before_merging_datasets": false,
+  "shuffle_merged_datasets": true,
+  "skip_prepare_dataset": false,
+  "special_tokens": {
+    "eos_token": "<|im_end|>",
+    "pad_token": "<|endoftext|>"
+  },
+  "streaming_multipack_buffer_size": 10000,
+  "strict": false,
+  "tensor_parallel_size": 1,
+  "tf32": true,
+  "tiled_mlp_use_original_mlp": true,
+  "tokenizer_config": "Qwen/Qwen3.5-9B",
+  "tokenizer_save_jinja_files": true,
+  "torch_dtype": "torch.bfloat16",
+  "train_on_inputs": false,
+  "trl": {
+    "async_prefetch": false,
+    "log_completions": false,
+    "mask_truncated_completions": false,
+    "ref_model_mixup_alpha": 0.9,
+    "ref_model_sync_steps": 64,
+    "replay_buffer_size": 0,
+    "replay_recompute_logps": true,
+    "reroll_max_groups": 1,
+    "reroll_start_fraction": 1.0,
+    "reward_num_workers": 1,
+    "scale_rewards": true,
+    "skip_zero_advantage_batches": true,
+    "sync_ref_model": false,
+    "use_data_producer": false,
+    "use_vllm": false,
+    "vllm_lora_sync": false,
+    "vllm_server_host": "0.0.0.0",
+    "vllm_server_port": 8000
+  },
+  "use_otel_metrics": false,
+  "use_ray": false,
+  "use_wandb": true,
+  "val_set_size": 0.0,
+  "vllm": {
+    "device": "auto",
+    "dtype": "auto",
+    "gpu_memory_utilization": 0.9,
+    "host": "0.0.0.0",
+    "port": 8000
+  },
+  "wandb_name": "philosophy-msm-aft-instruct-20260619-093540/msm",
+  "wandb_project": "why-gen",
+  "warmup_ratio": 0.05,
+  "weight_decay": 0.01,
+  "world_size": 1
+}
+[2026-06-19 09:36:59,259] [INFO] [axolotl.utils.data.shared] Loading prepared dataset from disk at /workspace/mats_project/data/.axolotl-prepared-cache/5ebaa067467de1cd17f91a288757e1fa...
+
+[2026-06-19 09:37:04,840] [INFO] [axolotl.utils.samplers.multipack] gather_len_batches: [5567]
+[2026-06-19 09:37:04,840] [INFO] [axolotl.utils.trainer] sample_packing_eff_est across ranks: [0.9188780465801357]
+[2026-06-19 09:37:04,841] [INFO] [axolotl.utils.data.sft] Maximum number of steps set at 1391
+[2026-06-19 09:37:09,731] [INFO] [axolotl.monkeypatch.attention.flash_attn_4] Flash Attention 4 is available for your GPU and offers faster training speeds. To enable: pip install flash-attn-4
+[2026-06-19 09:37:13,253] [INFO] [axolotl.monkeypatch.lora_kernels] Patched attention class with LoRA optims: Qwen3_5Attention
+[2026-06-19 09:37:13,266] [INFO] [axolotl.monkeypatch.models.qwen3_5.modeling] Applied Qwen3_5 packing patch (fla_causal_conv1d=available)
+[2026-06-19 09:37:13,266] [INFO] [axolotl.monkeypatch.models.qwen3_5.modeling] Applied Qwen3.5 VLM flash-attention patch (3-D MRoPE position_ids)
+[2026-06-19 09:37:13,269] [INFO] [axolotl.loaders.patch_manager] Applying multipack dataloader patch for sample packing...
+
+
+[transformers] You do not have `flash_attn` installed, using `kernels-community/flash-attn2` from the `kernels` library instead!
+
+[2026-06-19 09:37:17,288] [INFO] [axolotl.loaders.model] Converting modules to torch.bfloat16
+trainable params: 116,391,936 || all params: 9,526,205,680 || trainable%: 1.2218
+[2026-06-19 09:37:24,654] [INFO] [axolotl.train] Pre-saving adapter config to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm...
+[2026-06-19 09:37:24,658] [INFO] [axolotl.train] Pre-saving tokenizer to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm...
+[2026-06-19 09:37:24,807] [INFO] [axolotl.train] Pre-saving model config to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm...
+[2026-06-19 09:37:24,817] [INFO] [axolotl.train] Pre-saving processor to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm...
+[2026-06-19 09:37:24,970] [INFO] [axolotl.train] Starting trainer...
+[transformers] The tokenizer has new PAD/BOS/EOS tokens that differ from the model config and generation config. The model config and generation config were aligned accordingly, being updated with the tokenizer's values. Updated tokens: {'eos_token_id': 248046, 'pad_token_id': 248044}.
+[2026-06-19 09:37:30,721] [INFO] [axolotl.utils.samplers.multipack] gather_len_batches: [5571]
+wandb: Tracking run with wandb version 0.27.2
+wandb: W&B syncing is set to `offline` in this directory. Run `wandb online` or set WANDB_MODE=online to enable cloud syncing.
+wandb: Run data is saved locally in /workspace/wandb/wandb/offline-run-20260619_093731-cupn5mww
+wandb: WARNING Saving files without folders. If you want to preserve subdirectories pass base_path to wandb.save, i.e. wandb.save("/mnt/folder/file.h5", base_path="/mnt")
+wandb: WARNING Symlinked 1 file into the W&B run directory; call wandb.save again to sync new files.
+[2026-06-19 09:37:32,632] [INFO] [axolotl.utils.callbacks] The Axolotl config has been saved to the WandB run under files.
+{'loss': '1.524', 'grad_norm': '0.5396', 'learning_rate': '1.304e-05', 'ppl': '4.591', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1874', 'tokens/total': 327680, 'tokens/trainable': 320233, 'epoch': '0.00718'}
+{'loss': '1.423', 'grad_norm': '0.3517', 'learning_rate': '2.754e-05', 'ppl': '4.15', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1916', 'tokens/total': 655360, 'tokens/trainable': 636503, 'epoch': '0.01436'}
+{'loss': '1.356', 'grad_norm': '0.3654', 'learning_rate': '4.203e-05', 'ppl': '3.879', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1730', 'tokens/total': 983040, 'tokens/trainable': 951070, 'epoch': '0.02154'}
+{'loss': '1.282', 'grad_norm': '0.3798', 'learning_rate': '5.652e-05', 'ppl': '3.605', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1853', 'tokens/total': 1310720, 'tokens/trainable': 1267181, 'epoch': '0.02872'}
+{'loss': '1.221', 'grad_norm': '0.4241', 'learning_rate': '7.101e-05', 'ppl': '3.39', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1915', 'tokens/total': 1638400, 'tokens/trainable': 1584678, 'epoch': '0.0359'}
+{'loss': '1.177', 'grad_norm': '0.4011', 'learning_rate': '8.551e-05', 'ppl': '3.246', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1911', 'tokens/total': 1966080, 'tokens/trainable': 1895876, 'epoch': '0.04308'}
+{'loss': '1.135', 'grad_norm': '0.3944', 'learning_rate': '0.0001', 'ppl': '3.112', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1710', 'tokens/total': 2293760, 'tokens/trainable': 2211142, 'epoch': '0.05026'}
+{'loss': '1.128', 'grad_norm': '0.4057', 'learning_rate': '9.999e-05', 'ppl': '3.089', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1920', 'tokens/total': 2621440, 'tokens/trainable': 2523632, 'epoch': '0.05744'}
+{'loss': '1.102', 'grad_norm': '0.349', 'learning_rate': '9.994e-05', 'ppl': '3.009', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1774', 'tokens/total': 2949120, 'tokens/trainable': 2835508, 'epoch': '0.06462'}
+{'loss': '1.095', 'grad_norm': '0.3582', 'learning_rate': '9.987e-05', 'ppl': '2.988', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1920', 'tokens/total': 3276800, 'tokens/trainable': 3148695, 'epoch': '0.0718'}
+{'loss': '1.075', 'grad_norm': '0.3543', 'learning_rate': '9.977e-05', 'ppl': '2.93', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1842', 'tokens/total': 3604480, 'tokens/trainable': 3466044, 'epoch': '0.07898'}
+{'loss': '1.068', 'grad_norm': '0.3599', 'learning_rate': '9.965e-05', 'ppl': '2.909', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1905', 'tokens/total': 3932160, 'tokens/trainable': 3778534, 'epoch': '0.08616'}
+{'loss': '1.056', 'grad_norm': '0.3381', 'learning_rate': '9.949e-05', 'ppl': '2.874', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1802', 'tokens/total': 4259840, 'tokens/trainable': 4091431, 'epoch': '0.09334'}
+{'loss': '1.05', 'grad_norm': '0.3409', 'learning_rate': '9.931e-05', 'ppl': '2.859', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1901', 'tokens/total': 4587520, 'tokens/trainable': 4401406, 'epoch': '0.1005'}
+{'loss': '1.048', 'grad_norm': '0.3436', 'learning_rate': '9.91e-05', 'ppl': '2.853', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1892', 'tokens/total': 4915200, 'tokens/trainable': 4712142, 'epoch': '0.1077'}
+{'loss': '1.042', 'grad_norm': '0.3363', 'learning_rate': '9.886e-05', 'ppl': '2.834', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1698', 'tokens/total': 5242880, 'tokens/trainable': 5025694, 'epoch': '0.1149'}
+{'loss': '1.033', 'grad_norm': '0.3273', 'learning_rate': '9.859e-05', 'ppl': '2.811', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1721', 'tokens/total': 5570560, 'tokens/trainable': 5333753, 'epoch': '0.1221'}
+{'loss': '1.031', 'grad_norm': '0.344', 'learning_rate': '9.83e-05', 'ppl': '2.804', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1892', 'tokens/total': 5898240, 'tokens/trainable': 5641801, 'epoch': '0.1292'}
+{'loss': '1.011', 'grad_norm': '0.3328', 'learning_rate': '9.798e-05', 'ppl': '2.749', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1751', 'tokens/total': 6225920, 'tokens/trainable': 5951459, 'epoch': '0.1364'}
+{'loss': '1.015', 'grad_norm': '0.3229', 'learning_rate': '9.763e-05', 'ppl': '2.759', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1911', 'tokens/total': 6553600, 'tokens/trainable': 6261411, 'epoch': '0.1436'}
+{'loss': '1.011', 'grad_norm': '0.3312', 'learning_rate': '9.726e-05', 'ppl': '2.749', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1916', 'tokens/total': 6881280, 'tokens/trainable': 6565760, 'epoch': '0.1508'}
+{'loss': '1.015', 'grad_norm': '0.3399', 'learning_rate': '9.686e-05', 'ppl': '2.76', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1921', 'tokens/total': 7208960, 'tokens/trainable': 6876578, 'epoch': '0.158'}
+{'loss': '0.9883', 'grad_norm': '0.3277', 'learning_rate': '9.643e-05', 'ppl': '2.687', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1612', 'tokens/total': 7536640, 'tokens/trainable': 7181735, 'epoch': '0.1651'}
+{'loss': '1.014', 'grad_norm': '0.3264', 'learning_rate': '9.598e-05', 'ppl': '2.756', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1672', 'tokens/total': 7864320, 'tokens/trainable': 7490134, 'epoch': '0.1723'}
+{'loss': '0.9791', 'grad_norm': '0.3332', 'learning_rate': '9.55e-05', 'ppl': '2.662', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1664', 'tokens/total': 8192000, 'tokens/trainable': 7794665, 'epoch': '0.1795'}
+{'loss': '1.002', 'grad_norm': '0.3295', 'learning_rate': '9.499e-05', 'ppl': '2.724', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1645', 'tokens/total': 8519680, 'tokens/trainable': 8094467, 'epoch': '0.1867'}
+{'loss': '0.9751', 'grad_norm': '0.3087', 'learning_rate': '9.446e-05', 'ppl': '2.651', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1841', 'tokens/total': 8847360, 'tokens/trainable': 8395919, 'epoch': '0.1939'}
+{'loss': '0.9886', 'grad_norm': '0.3183', 'learning_rate': '9.39e-05', 'ppl': '2.688', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1571', 'tokens/total': 9175040, 'tokens/trainable': 8701190, 'epoch': '0.201'}
+{'loss': '0.9843', 'grad_norm': '0.321', 'learning_rate': '9.332e-05', 'ppl': '2.676', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1924', 'tokens/total': 9502720, 'tokens/trainable': 9007349, 'epoch': '0.2082'}
+{'loss': '0.9741', 'grad_norm': '0.3173', 'learning_rate': '9.272e-05', 'ppl': '2.649', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1909', 'tokens/total': 9830400, 'tokens/trainable': 9310722, 'epoch': '0.2154'}
+{'loss': '0.9714', 'grad_norm': '0.3275', 'learning_rate': '9.209e-05', 'ppl': '2.642', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1861', 'tokens/total': 10158080, 'tokens/trainable': 9620736, 'epoch': '0.2226'}
+{'loss': '0.9776', 'grad_norm': '0.3386', 'learning_rate': '9.143e-05', 'ppl': '2.658', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1906', 'tokens/total': 10485760, 'tokens/trainable': 9926004, 'epoch': '0.2298'}
+{'loss': '0.9853', 'grad_norm': '0.3339', 'learning_rate': '9.076e-05', 'ppl': '2.679', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1918', 'tokens/total': 10813440, 'tokens/trainable': 10230832, 'epoch': '0.2369'}
+{'loss': '0.9702', 'grad_norm': '0.3052', 'learning_rate': '9.006e-05', 'ppl': '2.638', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1902', 'tokens/total': 11141120, 'tokens/trainable': 10530921, 'epoch': '0.2441'}
+{'loss': '0.9703', 'grad_norm': '0.319', 'learning_rate': '8.933e-05', 'ppl': '2.639', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1793', 'tokens/total': 11468800, 'tokens/trainable': 10835657, 'epoch': '0.2513'}
+{'loss': '0.9586', 'grad_norm': '0.3331', 'learning_rate': '8.859e-05', 'ppl': '2.608', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1655', 'tokens/total': 11796480, 'tokens/trainable': 11132041, 'epoch': '0.2585'}
+{'loss': '0.9707', 'grad_norm': '0.3268', 'learning_rate': '8.782e-05', 'ppl': '2.64', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1888', 'tokens/total': 12124160, 'tokens/trainable': 11432692, 'epoch': '0.2657'}
+{'loss': '0.9769', 'grad_norm': '0.3501', 'learning_rate': '8.704e-05', 'ppl': '2.656', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1808', 'tokens/total': 12451840, 'tokens/trainable': 11734365, 'epoch': '0.2728'}
+{'loss': '0.9685', 'grad_norm': '0.3342', 'learning_rate': '8.623e-05', 'ppl': '2.634', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1657', 'tokens/total': 12779520, 'tokens/trainable': 12036905, 'epoch': '0.28'}
+{'loss': '0.9672', 'grad_norm': '0.3277', 'learning_rate': '8.54e-05', 'ppl': '2.63', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1541', 'tokens/total': 13107200, 'tokens/trainable': 12342375, 'epoch': '0.2872'}
+{'loss': '0.9673', 'grad_norm': '0.3244', 'learning_rate': '8.455e-05', 'ppl': '2.631', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1917', 'tokens/total': 13434880, 'tokens/trainable': 12642230, 'epoch': '0.2944'}
+{'loss': '0.968', 'grad_norm': '0.3343', 'learning_rate': '8.368e-05', 'ppl': '2.633', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1883', 'tokens/total': 13762560, 'tokens/trainable': 12943390, 'epoch': '0.3016'}
+{'loss': '0.9359', 'grad_norm': '0.3149', 'learning_rate': '8.279e-05', 'ppl': '2.55', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1923', 'tokens/total': 14090240, 'tokens/trainable': 13246463, 'epoch': '0.3087'}
+{'loss': '0.9469', 'grad_norm': '0.3205', 'learning_rate': '8.189e-05', 'ppl': '2.578', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1749', 'tokens/total': 14417920, 'tokens/trainable': 13545778, 'epoch': '0.3159'}
+{'loss': '0.9461', 'grad_norm': '0.3257', 'learning_rate': '8.096e-05', 'ppl': '2.576', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1865', 'tokens/total': 14745600, 'tokens/trainable': 13849693, 'epoch': '0.3231'}
+{'loss': '0.9401', 'grad_norm': '0.3253', 'learning_rate': '8.002e-05', 'ppl': '2.56', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1708', 'tokens/total': 15073280, 'tokens/trainable': 14150753, 'epoch': '0.3303'}
+{'loss': '0.9458', 'grad_norm': '0.332', 'learning_rate': '7.906e-05', 'ppl': '2.575', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1719', 'tokens/total': 15400960, 'tokens/trainable': 14446821, 'epoch': '0.3375'}
+{'loss': '0.9426', 'grad_norm': '0.3253', 'learning_rate': '7.809e-05', 'ppl': '2.567', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1824', 'tokens/total': 15728640, 'tokens/trainable': 14748728, 'epoch': '0.3446'}
+{'loss': '0.948', 'grad_norm': '0.3282', 'learning_rate': '7.71e-05', 'ppl': '2.58', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1883', 'tokens/total': 16056320, 'tokens/trainable': 15058308, 'epoch': '0.3518'}
+{'loss': '0.9544', 'grad_norm': '0.331', 'learning_rate': '7.609e-05', 'ppl': '2.597', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1557', 'tokens/total': 16384000, 'tokens/trainable': 15359225, 'epoch': '0.359'}
+{'loss': '0.9494', 'grad_norm': '0.3165', 'learning_rate': '7.507e-05', 'ppl': '2.584', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1889', 'tokens/total': 16711680, 'tokens/trainable': 15658552, 'epoch': '0.3662'}
+{'loss': '0.9306', 'grad_norm': '0.3154', 'learning_rate': '7.403e-05', 'ppl': '2.536', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1901', 'tokens/total': 17039360, 'tokens/trainable': 15962623, 'epoch': '0.3734'}
+{'loss': '0.958', 'grad_norm': '0.315', 'learning_rate': '7.298e-05', 'ppl': '2.606', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1874', 'tokens/total': 17367040, 'tokens/trainable': 16263771, 'epoch': '0.3805'}
+{'loss': '0.9309', 'grad_norm': '0.3317', 'learning_rate': '7.192e-05', 'ppl': '2.537', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1552', 'tokens/total': 17694720, 'tokens/trainable': 16569499, 'epoch': '0.3877'}
+{'loss': '0.9426', 'grad_norm': '0.3229', 'learning_rate': '7.085e-05', 'ppl': '2.567', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1709', 'tokens/total': 18022400, 'tokens/trainable': 16874608, 'epoch': '0.3949'}
+{'loss': '0.9153', 'grad_norm': '0.3347', 'learning_rate': '6.976e-05', 'ppl': '2.497', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1571', 'tokens/total': 18350080, 'tokens/trainable': 17173642, 'epoch': '0.4021'}
+{'loss': '0.9438', 'grad_norm': '0.3208', 'learning_rate': '6.867e-05', 'ppl': '2.57', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1684', 'tokens/total': 18677760, 'tokens/trainable': 17477160, 'epoch': '0.4093'}
+{'loss': '0.9421', 'grad_norm': '0.3099', 'learning_rate': '6.756e-05', 'ppl': '2.565', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1922', 'tokens/total': 19005440, 'tokens/trainable': 17783340, 'epoch': '0.4164'}
+{'loss': '0.9397', 'grad_norm': '0.3357', 'learning_rate': '6.644e-05', 'ppl': '2.559', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1914', 'tokens/total': 19333120, 'tokens/trainable': 18087280, 'epoch': '0.4236'}
+{'loss': '0.931', 'grad_norm': '0.3359', 'learning_rate': '6.532e-05', 'ppl': '2.537', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1668', 'tokens/total': 19660800, 'tokens/trainable': 18388320, 'epoch': '0.4308'}
+{'loss': '0.9394', 'grad_norm': '0.3362', 'learning_rate': '6.418e-05', 'ppl': '2.559', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1934', 'tokens/total': 19988480, 'tokens/trainable': 18691988, 'epoch': '0.438'}
+{'loss': '0.9118', 'grad_norm': '0.3281', 'learning_rate': '6.304e-05', 'ppl': '2.489', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1591', 'tokens/total': 20316160, 'tokens/trainable': 18992828, 'epoch': '0.4452'}
+{'loss': '0.9194', 'grad_norm': '0.3338', 'learning_rate': '6.189e-05', 'ppl': '2.508', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1643', 'tokens/total': 20643840, 'tokens/trainable': 19284462, 'epoch': '0.4523'}
+{'loss': '0.946', 'grad_norm': '0.3375', 'learning_rate': '6.073e-05', 'ppl': '2.575', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1823', 'tokens/total': 20971520, 'tokens/trainable': 19578686, 'epoch': '0.4595'}
+{'loss': '0.9343', 'grad_norm': '0.3336', 'learning_rate': '5.957e-05', 'ppl': '2.545', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1564', 'tokens/total': 21299200, 'tokens/trainable': 19877996, 'epoch': '0.4667'}
+{'loss': '0.9162', 'grad_norm': '0.3243', 'learning_rate': '5.84e-05', 'ppl': '2.5', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1756', 'tokens/total': 21626880, 'tokens/trainable': 20181408, 'epoch': '0.4739'}
+{'loss': '0.922', 'grad_norm': '0.338', 'learning_rate': '5.722e-05', 'ppl': '2.514', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1866', 'tokens/total': 21954560, 'tokens/trainable': 20479610, 'epoch': '0.4811'}
+{'loss': '0.9004', 'grad_norm': '0.3505', 'learning_rate': '5.604e-05', 'ppl': '2.461', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1667', 'tokens/total': 22282240, 'tokens/trainable': 20779428, 'epoch': '0.4882'}
+{'loss': '0.9177', 'grad_norm': '0.3201', 'learning_rate': '5.486e-05', 'ppl': '2.504', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1684', 'tokens/total': 22609920, 'tokens/trainable': 21080584, 'epoch': '0.4954'}
+[2026-06-19 10:26:57,012] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-696
+{'loss': '0.9206', 'grad_norm': '0.3295', 'learning_rate': '5.368e-05', 'ppl': '2.511', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.33', 'tokens/train_per_sec_per_gpu': '1625', 'tokens/total': 22937600, 'tokens/trainable': 21374452, 'epoch': '0.5026'}
+{'loss': '0.9292', 'grad_norm': '0.3516', 'learning_rate': '5.249e-05', 'ppl': '2.532', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1724', 'tokens/total': 23265280, 'tokens/trainable': 21671040, 'epoch': '0.5098'}
+{'loss': '0.9256', 'grad_norm': '0.3382', 'learning_rate': '5.131e-05', 'ppl': '2.523', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1583', 'tokens/total': 23592960, 'tokens/trainable': 21973320, 'epoch': '0.517'}
+{'loss': '0.9128', 'grad_norm': '0.344', 'learning_rate': '5.012e-05', 'ppl': '2.491', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1891', 'tokens/total': 23920640, 'tokens/trainable': 22272124, 'epoch': '0.5241'}
+{'loss': '0.9066', 'grad_norm': '0.3214', 'learning_rate': '4.893e-05', 'ppl': '2.476', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1903', 'tokens/total': 24248320, 'tokens/trainable': 22574372, 'epoch': '0.5313'}
+{'loss': '0.9222', 'grad_norm': '0.3292', 'learning_rate': '4.774e-05', 'ppl': '2.515', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1922', 'tokens/total': 24576000, 'tokens/trainable': 22873552, 'epoch': '0.5385'}
+{'loss': '0.9089', 'grad_norm': '0.3454', 'learning_rate': '4.656e-05', 'ppl': '2.482', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1687', 'tokens/total': 24903680, 'tokens/trainable': 23166824, 'epoch': '0.5457'}
+{'loss': '0.9104', 'grad_norm': '0.3266', 'learning_rate': '4.537e-05', 'ppl': '2.485', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1607', 'tokens/total': 25231360, 'tokens/trainable': 23469360, 'epoch': '0.5529'}
+{'loss': '0.9064', 'grad_norm': '0.3365', 'learning_rate': '4.419e-05', 'ppl': '2.475', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1622', 'tokens/total': 25559040, 'tokens/trainable': 23768178, 'epoch': '0.56'}
+{'loss': '0.9193', 'grad_norm': '0.3306', 'learning_rate': '4.301e-05', 'ppl': '2.508', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1925', 'tokens/total': 25886720, 'tokens/trainable': 24073610, 'epoch': '0.5672'}
+{'loss': '0.9219', 'grad_norm': '0.3375', 'learning_rate': '4.184e-05', 'ppl': '2.514', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1706', 'tokens/total': 26214400, 'tokens/trainable': 24370876, 'epoch': '0.5744'}
+{'loss': '0.9025', 'grad_norm': '0.3349', 'learning_rate': '4.067e-05', 'ppl': '2.466', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1923', 'tokens/total': 26542080, 'tokens/trainable': 24666098, 'epoch': '0.5816'}
+{'loss': '0.9262', 'grad_norm': '0.3488', 'learning_rate': '3.95e-05', 'ppl': '2.525', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1549', 'tokens/total': 26869760, 'tokens/trainable': 24964708, 'epoch': '0.5888'}
+{'loss': '0.9231', 'grad_norm': '0.3443', 'learning_rate': '3.835e-05', 'ppl': '2.517', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1727', 'tokens/total': 27197440, 'tokens/trainable': 25261144, 'epoch': '0.5959'}
+{'loss': '0.9043', 'grad_norm': '0.3234', 'learning_rate': '3.719e-05', 'ppl': '2.47', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1719', 'tokens/total': 27525120, 'tokens/trainable': 25568364, 'epoch': '0.6031'}
+{'loss': '0.9137', 'grad_norm': '0.3299', 'learning_rate': '3.605e-05', 'ppl': '2.494', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1763', 'tokens/total': 27852800, 'tokens/trainable': 25862664, 'epoch': '0.6103'}
+{'loss': '0.9102', 'grad_norm': '0.3389', 'learning_rate': '3.491e-05', 'ppl': '2.485', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1917', 'tokens/total': 28180480, 'tokens/trainable': 26164610, 'epoch': '0.6175'}
+{'loss': '0.8983', 'grad_norm': '0.328', 'learning_rate': '3.378e-05', 'ppl': '2.455', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1878', 'tokens/total': 28508160, 'tokens/trainable': 26456168, 'epoch': '0.6247'}
+{'loss': '0.9212', 'grad_norm': '0.3245', 'learning_rate': '3.266e-05', 'ppl': '2.512', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1841', 'tokens/total': 28835840, 'tokens/trainable': 26760032, 'epoch': '0.6318'}
+{'loss': '0.9042', 'grad_norm': '0.3343', 'learning_rate': '3.155e-05', 'ppl': '2.47', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1865', 'tokens/total': 29163520, 'tokens/trainable': 27061136, 'epoch': '0.639'}
+{'loss': '0.8849', 'grad_norm': '0.3452', 'learning_rate': '3.045e-05', 'ppl': '2.423', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1529', 'tokens/total': 29491200, 'tokens/trainable': 27358870, 'epoch': '0.6462'}
+{'loss': '0.8996', 'grad_norm': '0.3349', 'learning_rate': '2.937e-05', 'ppl': '2.459', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1682', 'tokens/total': 29818880, 'tokens/trainable': 27657278, 'epoch': '0.6534'}
+{'loss': '0.8885', 'grad_norm': '0.3268', 'learning_rate': '2.829e-05', 'ppl': '2.432', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1789', 'tokens/total': 30146560, 'tokens/trainable': 27955380, 'epoch': '0.6606'}
+{'loss': '0.9002', 'grad_norm': '0.3178', 'learning_rate': '2.723e-05', 'ppl': '2.46', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1880', 'tokens/total': 30474240, 'tokens/trainable': 28256136, 'epoch': '0.6677'}
+{'loss': '0.8904', 'grad_norm': '0.3554', 'learning_rate': '2.618e-05', 'ppl': '2.436', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1715', 'tokens/total': 30801920, 'tokens/trainable': 28561112, 'epoch': '0.6749'}
+{'loss': '0.9151', 'grad_norm': '0.3344', 'learning_rate': '2.514e-05', 'ppl': '2.497', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1805', 'tokens/total': 31129600, 'tokens/trainable': 28852180, 'epoch': '0.6821'}
+{'loss': '0.9022', 'grad_norm': '0.3476', 'learning_rate': '2.411e-05', 'ppl': '2.465', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1581', 'tokens/total': 31457280, 'tokens/trainable': 29149028, 'epoch': '0.6893'}
+{'loss': '0.889', 'grad_norm': '0.3328', 'learning_rate': '2.31e-05', 'ppl': '2.433', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1732', 'tokens/total': 31784960, 'tokens/trainable': 29444158, 'epoch': '0.6965'}
+{'loss': '0.9074', 'grad_norm': '0.3311', 'learning_rate': '2.211e-05', 'ppl': '2.478', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1920', 'tokens/total': 32112640, 'tokens/trainable': 29737064, 'epoch': '0.7036'}
+{'loss': '0.9037', 'grad_norm': '0.3371', 'learning_rate': '2.113e-05', 'ppl': '2.469', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1873', 'tokens/total': 32440320, 'tokens/trainable': 30032704, 'epoch': '0.7108'}
+{'loss': '0.8964', 'grad_norm': '0.3424', 'learning_rate': '2.017e-05', 'ppl': '2.451', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1686', 'tokens/total': 32768000, 'tokens/trainable': 30328108, 'epoch': '0.718'}
+{'loss': '0.9033', 'grad_norm': '0.3354', 'learning_rate': '1.923e-05', 'ppl': '2.468', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1798', 'tokens/total': 33095680, 'tokens/trainable': 30627696, 'epoch': '0.7252'}
+{'loss': '0.8936', 'grad_norm': '0.3397', 'learning_rate': '1.83e-05', 'ppl': '2.444', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1913', 'tokens/total': 33423360, 'tokens/trainable': 30925488, 'epoch': '0.7324'}
+{'loss': '0.9025', 'grad_norm': '0.335', 'learning_rate': '1.739e-05', 'ppl': '2.466', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1897', 'tokens/total': 33751040, 'tokens/trainable': 31223512, 'epoch': '0.7395'}
+{'loss': '0.8977', 'grad_norm': '0.3316', 'learning_rate': '1.65e-05', 'ppl': '2.454', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1904', 'tokens/total': 34078720, 'tokens/trainable': 31524356, 'epoch': '0.7467'}
+{'loss': '0.9036', 'grad_norm': '0.3448', 'learning_rate': '1.562e-05', 'ppl': '2.468', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1733', 'tokens/total': 34406400, 'tokens/trainable': 31817444, 'epoch': '0.7539'}
+{'loss': '0.8911', 'grad_norm': '0.3392', 'learning_rate': '1.477e-05', 'ppl': '2.438', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1658', 'tokens/total': 34734080, 'tokens/trainable': 32115804, 'epoch': '0.7611'}
+{'loss': '0.8989', 'grad_norm': '0.3369', 'learning_rate': '1.394e-05', 'ppl': '2.457', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1497', 'tokens/total': 35061760, 'tokens/trainable': 32409324, 'epoch': '0.7683'}
+{'loss': '0.909', 'grad_norm': '0.3363', 'learning_rate': '1.312e-05', 'ppl': '2.482', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1817', 'tokens/total': 35389440, 'tokens/trainable': 32699570, 'epoch': '0.7754'}
+{'loss': '0.9038', 'grad_norm': '0.3477', 'learning_rate': '1.233e-05', 'ppl': '2.469', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1875', 'tokens/total': 35717120, 'tokens/trainable': 32993832, 'epoch': '0.7826'}
+{'loss': '0.8945', 'grad_norm': '0.3479', 'learning_rate': '1.156e-05', 'ppl': '2.446', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1629', 'tokens/total': 36044800, 'tokens/trainable': 33296596, 'epoch': '0.7898'}
+{'loss': '0.8775', 'grad_norm': '0.345', 'learning_rate': '1.081e-05', 'ppl': '2.405', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1630', 'tokens/total': 36372480, 'tokens/trainable': 33597972, 'epoch': '0.797'}
+{'loss': '0.8891', 'grad_norm': '0.3517', 'learning_rate': '1.009e-05', 'ppl': '2.433', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1759', 'tokens/total': 36700160, 'tokens/trainable': 33893216, 'epoch': '0.8042'}
+{'loss': '0.8929', 'grad_norm': '0.3427', 'learning_rate': '9.382e-06', 'ppl': '2.442', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1547', 'tokens/total': 37027840, 'tokens/trainable': 34185752, 'epoch': '0.8113'}
+{'loss': '0.8969', 'grad_norm': '0.3324', 'learning_rate': '8.701e-06', 'ppl': '2.452', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1916', 'tokens/total': 37355520, 'tokens/trainable': 34477912, 'epoch': '0.8185'}
+{'loss': '0.9027', 'grad_norm': '0.3528', 'learning_rate': '8.043e-06', 'ppl': '2.466', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1910', 'tokens/total': 37683200, 'tokens/trainable': 34776324, 'epoch': '0.8257'}
+{'loss': '0.8868', 'grad_norm': '0.3396', 'learning_rate': '7.408e-06', 'ppl': '2.427', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1521', 'tokens/total': 38010880, 'tokens/trainable': 35068108, 'epoch': '0.8329'}
+{'loss': '0.883', 'grad_norm': '0.3502', 'learning_rate': '6.798e-06', 'ppl': '2.418', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1582', 'tokens/total': 38338560, 'tokens/trainable': 35365276, 'epoch': '0.8401'}
+{'loss': '0.905', 'grad_norm': '0.3361', 'learning_rate': '6.212e-06', 'ppl': '2.472', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1886', 'tokens/total': 38666240, 'tokens/trainable': 35662984, 'epoch': '0.8472'}
+{'loss': '0.8928', 'grad_norm': '0.3509', 'learning_rate': '5.651e-06', 'ppl': '2.442', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1744', 'tokens/total': 38993920, 'tokens/trainable': 35961584, 'epoch': '0.8544'}
+{'loss': '0.899', 'grad_norm': '0.3267', 'learning_rate': '5.115e-06', 'ppl': '2.457', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1686', 'tokens/total': 39321600, 'tokens/trainable': 36257436, 'epoch': '0.8616'}
+{'loss': '0.8924', 'grad_norm': '0.3435', 'learning_rate': '4.604e-06', 'ppl': '2.441', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1594', 'tokens/total': 39649280, 'tokens/trainable': 36553308, 'epoch': '0.8688'}
+{'loss': '0.8807', 'grad_norm': '0.3505', 'learning_rate': '4.119e-06', 'ppl': '2.413', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1678', 'tokens/total': 39976960, 'tokens/trainable': 36845968, 'epoch': '0.876'}
+{'loss': '0.903', 'grad_norm': '0.357', 'learning_rate': '3.66e-06', 'ppl': '2.467', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1623', 'tokens/total': 40304640, 'tokens/trainable': 37137272, 'epoch': '0.8831'}
+{'loss': '0.8935', 'grad_norm': '0.3514', 'learning_rate': '3.227e-06', 'ppl': '2.444', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1781', 'tokens/total': 40632320, 'tokens/trainable': 37433968, 'epoch': '0.8903'}
+{'loss': '0.9011', 'grad_norm': '0.3621', 'learning_rate': '2.82e-06', 'ppl': '2.462', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1514', 'tokens/total': 40960000, 'tokens/trainable': 37727884, 'epoch': '0.8975'}
+{'loss': '0.8976', 'grad_norm': '0.3414', 'learning_rate': '2.44e-06', 'ppl': '2.454', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1771', 'tokens/total': 41287680, 'tokens/trainable': 38024668, 'epoch': '0.9047'}
+{'loss': '0.8833', 'grad_norm': '0.3337', 'learning_rate': '2.087e-06', 'ppl': '2.419', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1817', 'tokens/total': 41615360, 'tokens/trainable': 38318496, 'epoch': '0.9119'}
+{'loss': '0.894', 'grad_norm': '0.3441', 'learning_rate': '1.761e-06', 'ppl': '2.445', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1903', 'tokens/total': 41943040, 'tokens/trainable': 38611004, 'epoch': '0.919'}
+{'loss': '0.8931', 'grad_norm': '0.3395', 'learning_rate': '1.462e-06', 'ppl': '2.443', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1931', 'tokens/total': 42270720, 'tokens/trainable': 38907484, 'epoch': '0.9262'}
+{'loss': '0.8965', 'grad_norm': '0.3524', 'learning_rate': '1.19e-06', 'ppl': '2.451', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1569', 'tokens/total': 42598400, 'tokens/trainable': 39203216, 'epoch': '0.9334'}
+{'loss': '0.8888', 'grad_norm': '0.3363', 'learning_rate': '9.463e-07', 'ppl': '2.432', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1904', 'tokens/total': 42926080, 'tokens/trainable': 39501840, 'epoch': '0.9406'}
+{'loss': '0.8847', 'grad_norm': '0.3354', 'learning_rate': '7.301e-07', 'ppl': '2.422', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1896', 'tokens/total': 43253760, 'tokens/trainable': 39792848, 'epoch': '0.9478'}
+{'loss': '0.8907', 'grad_norm': '0.337', 'learning_rate': '5.417e-07', 'ppl': '2.437', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1630', 'tokens/total': 43581440, 'tokens/trainable': 40079552, 'epoch': '0.9549'}
+{'loss': '0.8961', 'grad_norm': '0.3338', 'learning_rate': '3.813e-07', 'ppl': '2.45', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1879', 'tokens/total': 43909120, 'tokens/trainable': 40373720, 'epoch': '0.9621'}
+{'loss': '0.9092', 'grad_norm': '0.3484', 'learning_rate': '2.488e-07', 'ppl': '2.482', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1640', 'tokens/total': 44236800, 'tokens/trainable': 40660548, 'epoch': '0.9693'}
+{'loss': '0.8936', 'grad_norm': '0.3589', 'learning_rate': '1.445e-07', 'ppl': '2.444', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1485', 'tokens/total': 44564480, 'tokens/trainable': 40953568, 'epoch': '0.9765'}
+{'loss': '0.8941', 'grad_norm': '0.3604', 'learning_rate': '6.832e-08', 'ppl': '2.445', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1676', 'tokens/total': 44892160, 'tokens/trainable': 41241024, 'epoch': '0.9837'}
+{'loss': '0.8838', 'grad_norm': '0.348', 'learning_rate': '2.033e-08', 'ppl': '2.42', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1896', 'tokens/total': 45219840, 'tokens/trainable': 41533044, 'epoch': '0.9908'}
+{'loss': '0.8873', 'grad_norm': '0.3344', 'learning_rate': '5.647e-10', 'ppl': '2.429', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'tokens/train_per_sec_per_gpu': '1896', 'tokens/total': 45547520, 'tokens/trainable': 41824232, 'epoch': '0.998'}
+[2026-06-19 11:16:07,648] [INFO] [axolotl.core.trainers.base] Saving model checkpoint to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm/checkpoint-1391
+{'train_runtime': '5918', 'train_samples_per_second': '0.94', 'train_steps_per_second': '0.235', 'train_loss': '0.955', 'memory/max_active (GiB)': '44.19', 'memory/max_allocated (GiB)': '44.19', 'memory/device_reserved (GiB)': '48.31', 'epoch': '0.9987', 'tokens/train_per_sec_per_gpu': '1578'}
+
+[2026-06-19 11:16:09,277] [INFO] [axolotl.train] Training completed! Saving trained model to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm.
+[2026-06-19 11:16:10,082] [INFO] [axolotl.train] Model successfully saved to /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/msm
+wandb: 
+wandb: You can sync this run to the cloud by running:
+wandb: wandb sync /workspace/wandb/wandb/offline-run-20260619_093731-cupn5mww
+wandb: Find logs at: ../../../wandb/wandb/offline-run-20260619_093731-cupn5mww/logs
+

qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/logs/orchestrator.log

@@ -0,0 +1,10 @@
+2026-06-19 09:35:40,768 why_gen.train INFO run dir: /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540
+2026-06-19 09:35:40,776 why_gen.train INFO emitted /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/axolotl/msm.yaml
+2026-06-19 09:35:40,796 why_gen.train INFO emitted /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/axolotl/aft.yaml
+2026-06-19 09:35:40,799 why_gen.train INFO stage msm starting; trainer log: /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/logs/msm.log
+2026-06-19 09:35:40,799 why_gen.train INFO trainer cmd: axolotl train /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/axolotl/msm.yaml
+2026-06-19 11:16:16,605 why_gen.train INFO stage msm finished: exit=0 in 100.6 min
+2026-06-19 11:16:16,609 why_gen.train INFO stage aft starting; trainer log: /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/logs/aft.log
+2026-06-19 11:16:16,609 why_gen.train INFO trainer cmd: axolotl train /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/axolotl/aft.yaml
+2026-06-19 11:45:25,936 why_gen.train INFO stage aft finished: exit=0 in 29.2 min
+2026-06-19 11:45:25,947 why_gen.train INFO run philosophy-msm-aft-instruct-20260619-093540 complete. Next: python -m why_gen.evaluate --run-dir /workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540

qwen35_9b_exp2_overnight_state/base_msm.path

@@ -0,0 +1 @@
+/workspace/mats_project/data/runs/qwen35_9b_exp2_base/msm-only-base-20260619-075037/checkpoints/msm

qwen35_9b_exp2_overnight_state/inst_aft.path

@@ -0,0 +1 @@
+/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/aft-cot-only-instruct-20260619-093626/checkpoints/aft

qwen35_9b_exp2_overnight_state/inst_aft_msm.path

@@ -0,0 +1 @@
+/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-aft-msm-instruct-20260619-100857/checkpoints/msm

qwen35_9b_exp2_overnight_state/inst_msm.path

@@ -0,0 +1 @@
+/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/msm-only-instruct-20260619-075038/checkpoints/msm

qwen35_9b_exp2_overnight_state/inst_msm_aft.path

@@ -0,0 +1 @@
+/workspace/mats_project/data/runs/qwen35_9b_exp2_instruct/philosophy-msm-aft-instruct-20260619-093540/checkpoints/aft

qwen_swap/a1prime-bf16/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "/workspace/mats_project/data/runs/qwen_swap/merged-b-m1",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": null,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "down_proj",
+    "v_proj"
+  ],
+  "target_parameters": [],
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

qwen_swap/a1prime-clean-20260616-023928/a1prime.yaml

@@ -0,0 +1,64 @@
+sample_packing: true
+base_model: /workspace/mats_project/data/runs/qwen_swap/merged-b-m1
+load_in_8bit: false
+special_tokens:
+  pad_token: <|endoftext|>
+  eos_token: <|im_end|>
+adapter: lora
+lora_r: 64
+lora_alpha: 128
+lora_target_modules:
+- q_proj
+- k_proj
+- v_proj
+- o_proj
+- gate_proj
+- up_proj
+- down_proj
+lora_dropout: 0
+lora_mlp_kernel: false
+lora_qkv_kernel: false
+lora_o_kernel: false
+micro_batch_size: 1
+gradient_accumulation_steps: 16
+gradient_checkpointing: true
+learning_rate: 1e-4
+lr_scheduler: cosine
+warmup_ratio: 0.05
+weight_decay: 0.01
+max_grad_norm: 1.0
+optimizer: adamw_torch_fused
+saves_per_epoch: 2
+logging_steps: 10
+debug: true
+output_dir: /workspace/mats_project/data/runs/qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime
+auto_resume_from_checkpoints: true
+use_wandb: true
+wandb_project: why-gen
+bf16: true
+tf32: true
+flash_attention: true
+chat_template: jinja
+chat_template_jinja: '{%- if messages[0][''role''] != ''system'' %}{{- ''<|im_start|>system
+
+  You are a helpful assistant.<|im_end|>
+
+  '' }}{%- endif %}{%- set ns = namespace(last_query_index=messages|length - 1) %}{%- for message in messages %}{%- if message.role == "user" %}{%- set ns.last_query_index = loop.index0 %}{%- endif %}{%- endfor %}{%- for message in messages %}{%- if message.content is string %}{%- set content = message.content %}{%- else %}{%- set content = '''' %}{%- endif %}{%- if message.role == "assistant" %}{%- set reasoning_content = '''' %}{%- if message.reasoning_content is string %}{%- set reasoning_content = message.reasoning_content %}{%- else %}{%- if ''</think>'' in content %}{%- set reasoning_content = content.split(''</think>'')[0].rstrip(''\n'').split(''<think>'')[-1].lstrip(''\n'') %}{%- set content = content.split(''</think>'')[-1].lstrip(''\n'') %}{%- endif %}{%- endif %}{%- if loop.index0 > ns.last_query_index %}{%- if loop.last or (not loop.last and reasoning_content) %}{{- ''<|im_start|>'' + message.role + ''\n<think>\n'' + reasoning_content.strip(''\n'') + ''\n</think>\n\n'' + content.lstrip(''\n'') }}{%- else %}{{- ''<|im_start|>'' + message.role + ''\n'' + content }}{%- endif %}{%- else %}{{- ''<|im_start|>'' + message.role + ''\n'' + content }}{%- endif %}{{- ''<|im_end|>\n'' }}{%- else %}{{- ''<|im_start|>'' + message.role + ''\n'' + content + ''<|im_end|>\n'' }}{%- endif %}{%- endfor %}{%- if add_generation_prompt %}{{- ''<|im_start|>assistant\n'' }}{%- if enable_thinking is defined and enable_thinking is false %}{{- ''<think>\n\n</think>\n\n'' }}{%- endif %}{%- endif %}'
+dataset_prepared_path: /root/.axolotl-prepared-cache
+datasets:
+- path: /workspace/mats_project/data/msm/aft-cot-qwen3-philosophy-spec.jsonl
+  type: chat_template
+  field_messages: messages
+- path: /workspace/mats_project/data/built/it-mix-table2.jsonl
+  type: chat_template
+  field_messages: messages
+num_epochs: 1
+wandb_name: a1prime-clean/aft
+sequence_len: 8192
+activation_offloading: true
+plugins:
+- axolotl.integrations.liger.LigerPlugin
+liger_rope: true
+liger_rms_norm: true
+liger_glu_activation: true
+liger_fused_linear_cross_entropy: true

qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime/README.md

@@ -0,0 +1,137 @@
+---
+library_name: peft
+tags:
+- axolotl
+- base_model:adapter:/workspace/mats_project/data/runs/qwen_swap/merged-b-m1
+- lora
+- transformers
+datasets:
+- /workspace/mats_project/data/msm/aft-cot-qwen3-philosophy-spec.jsonl
+- /workspace/mats_project/data/built/it-mix-table2.jsonl
+base_model: /workspace/mats_project/data/runs/qwen_swap/merged-b-m1
+pipeline_tag: text-generation
+model-index:
+- name: workspace/mats_project/data/runs/qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
+<details><summary>See axolotl config</summary>
+
+axolotl version: `0.12.2`
+```yaml
+sample_packing: true
+base_model: /workspace/mats_project/data/runs/qwen_swap/merged-b-m1
+load_in_8bit: false
+special_tokens:
+  pad_token: <|endoftext|>
+  eos_token: <|im_end|>
+adapter: lora
+lora_r: 64
+lora_alpha: 128
+lora_target_modules:
+- q_proj
+- k_proj
+- v_proj
+- o_proj
+- gate_proj
+- up_proj
+- down_proj
+lora_dropout: 0
+lora_mlp_kernel: false
+lora_qkv_kernel: false
+lora_o_kernel: false
+micro_batch_size: 1
+gradient_accumulation_steps: 16
+gradient_checkpointing: true
+learning_rate: 1e-4
+lr_scheduler: cosine
+warmup_ratio: 0.05
+weight_decay: 0.01
+max_grad_norm: 1.0
+optimizer: adamw_torch_fused
+saves_per_epoch: 2
+logging_steps: 10
+debug: true
+output_dir: /workspace/mats_project/data/runs/qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime
+auto_resume_from_checkpoints: true
+use_wandb: true
+wandb_project: why-gen
+bf16: true
+tf32: true
+flash_attention: true
+chat_template: jinja
+chat_template_jinja: '{%- if messages[0][''role''] != ''system'' %}{{- ''<|im_start|>system
+
+  You are a helpful assistant.<|im_end|>
+
+  '' }}{%- endif %}{%- set ns = namespace(last_query_index=messages|length - 1) %}{%- for message in messages %}{%- if message.role == "user" %}{%- set ns.last_query_index = loop.index0 %}{%- endif %}{%- endfor %}{%- for message in messages %}{%- if message.content is string %}{%- set content = message.content %}{%- else %}{%- set content = '''' %}{%- endif %}{%- if message.role == "assistant" %}{%- set reasoning_content = '''' %}{%- if message.reasoning_content is string %}{%- set reasoning_content = message.reasoning_content %}{%- else %}{%- if ''</think>'' in content %}{%- set reasoning_content = content.split(''</think>'')[0].rstrip(''\n'').split(''<think>'')[-1].lstrip(''\n'') %}{%- set content = content.split(''</think>'')[-1].lstrip(''\n'') %}{%- endif %}{%- endif %}{%- if loop.index0 > ns.last_query_index %}{%- if loop.last or (not loop.last and reasoning_content) %}{{- ''<|im_start|>'' + message.role + ''\n<think>\n'' + reasoning_content.strip(''\n'') + ''\n</think>\n\n'' + content.lstrip(''\n'') }}{%- else %}{{- ''<|im_start|>'' + message.role + ''\n'' + content }}{%- endif %}{%- else %}{{- ''<|im_start|>'' + message.role + ''\n'' + content }}{%- endif %}{{- ''<|im_end|>\n'' }}{%- else %}{{- ''<|im_start|>'' + message.role + ''\n'' + content + ''<|im_end|>\n'' }}{%- endif %}{%- endfor %}{%- if add_generation_prompt %}{{- ''<|im_start|>assistant\n'' }}{%- if enable_thinking is defined and enable_thinking is false %}{{- ''<think>\n\n</think>\n\n'' }}{%- endif %}{%- endif %}'
+dataset_prepared_path: /root/.axolotl-prepared-cache
+datasets:
+- path: /workspace/mats_project/data/msm/aft-cot-qwen3-philosophy-spec.jsonl
+  type: chat_template
+  field_messages: messages
+- path: /workspace/mats_project/data/built/it-mix-table2.jsonl
+  type: chat_template
+  field_messages: messages
+num_epochs: 1
+wandb_name: a1prime-clean/aft
+sequence_len: 8192
+activation_offloading: true
+plugins:
+- axolotl.integrations.liger.LigerPlugin
+liger_rope: true
+liger_rms_norm: true
+liger_glu_activation: true
+liger_fused_linear_cross_entropy: true
+
+```
+
+</details><br>
+
+# workspace/mats_project/data/runs/qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime
+
+This model was trained from scratch on the /workspace/mats_project/data/msm/aft-cot-qwen3-philosophy-spec.jsonl and the /workspace/mats_project/data/built/it-mix-table2.jsonl datasets.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 0.0001
+- train_batch_size: 1
+- eval_batch_size: 1
+- seed: 42
+- gradient_accumulation_steps: 16
+- total_train_batch_size: 16
+- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_steps: 4
+- training_steps: 98
+
+### Training results
+
+
+
+### Framework versions
+
+- PEFT 0.17.0
+- Transformers 4.55.2
+- Pytorch 2.6.0+cu124
+- Datasets 4.0.0
+- Tokenizers 0.21.4

qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "/workspace/mats_project/data/runs/qwen_swap/merged-b-m1",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": null,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "down_proj",
+    "v_proj"
+  ],
+  "target_parameters": [],
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime/added_tokens.json

@@ -0,0 +1,28 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime/chat_template.jinja

@@ -0,0 +1,3 @@
+{%- if messages[0]['role'] != 'system' %}{{- '<|im_start|>system
+You are a helpful assistant.<|im_end|>
+' }}{%- endif %}{%- set ns = namespace(last_query_index=messages|length - 1) %}{%- for message in messages %}{%- if message.role == "user" %}{%- set ns.last_query_index = loop.index0 %}{%- endif %}{%- endfor %}{%- for message in messages %}{%- if message.content is string %}{%- set content = message.content %}{%- else %}{%- set content = '' %}{%- endif %}{%- if message.role == "assistant" %}{%- set reasoning_content = '' %}{%- if message.reasoning_content is string %}{%- set reasoning_content = message.reasoning_content %}{%- else %}{%- if '</think>' in content %}{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}{%- set content = content.split('</think>')[-1].lstrip('\n') %}{%- endif %}{%- endif %}{%- if loop.index0 > ns.last_query_index %}{%- if loop.last or (not loop.last and reasoning_content) %}{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}{%- else %}{{- '<|im_start|>' + message.role + '\n' + content }}{%- endif %}{%- else %}{{- '<|im_start|>' + message.role + '\n' + content }}{%- endif %}{{- '<|im_end|>\n' }}{%- else %}{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>\n' }}{%- endif %}{%- endfor %}{%- if add_generation_prompt %}{{- '<|im_start|>assistant\n' }}{%- if enable_thinking is defined and enable_thinking is false %}{{- '<think>\n\n</think>\n\n' }}{%- endif %}{%- endif %}

qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime/checkpoint-49/README.md

@@ -0,0 +1,208 @@
+---
+base_model: /workspace/mats_project/data/runs/qwen_swap/merged-b-m1
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- axolotl
+- base_model:adapter:/workspace/mats_project/data/runs/qwen_swap/merged-b-m1
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.17.0

qwen_swap/a1prime-clean-20260616-023928/checkpoints/a1prime/checkpoint-49/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "/workspace/mats_project/data/runs/qwen_swap/merged-b-m1",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": null,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "down_proj",
+    "v_proj"
+  ],
+  "target_parameters": [],
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}