Instructions to use dphn/dolphin-2.9.1-dbrx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use dphn/dolphin-2.9.1-dbrx with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="dphn/dolphin-2.9.1-dbrx", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("dphn/dolphin-2.9.1-dbrx", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("dphn/dolphin-2.9.1-dbrx", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use dphn/dolphin-2.9.1-dbrx with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "dphn/dolphin-2.9.1-dbrx"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "dphn/dolphin-2.9.1-dbrx",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/dphn/dolphin-2.9.1-dbrx

SGLang

How to use dphn/dolphin-2.9.1-dbrx with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "dphn/dolphin-2.9.1-dbrx" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "dphn/dolphin-2.9.1-dbrx",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "dphn/dolphin-2.9.1-dbrx" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "dphn/dolphin-2.9.1-dbrx",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use dphn/dolphin-2.9.1-dbrx with Docker Model Runner:
```
docker model run hf.co/dphn/dolphin-2.9.1-dbrx
```

Dolphin 2.9.1 DBRX 🐬

Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations

Discord: https://discord.gg/cognitivecomputations

Our appreciation for the sponsors of Dolphin 2.9.1:

Crusoe Cloud - provided excellent on-demand 8xH100 node

This model is based on databricks/dbrx-base, and is governed by databricks-open-model-license

The base model has 32k context, and the full-weight fine-tuning was with 4k sequence length.

This model was trained FFT on parameters selected by Laser Scanner, using ChatML prompt template format.

example:

<|im_start|>system
You are Dolphin, a helpful AI assistant.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

Dolphin-2.9.1 has a variety of instruction, conversational, and coding skills. It also has initial agentic abilities and supports function calling.

Dolphin is uncensored. We have filtered the dataset to remove alignment and bias. This makes the model more compliant. You are advised to implement your own alignment layer before exposing the model as a service. It will be highly compliant with any requests, even unethical ones. Please read my blog post about uncensored models. https://erichartford.com/uncensored-models You are responsible for any content you create using this model. Enjoy responsibly.

Dolphin is licensed according to Meta's Llama license. We grant permission for any use, including commercial, that falls within accordance with Meta's Llama-3 license. Dolphin was trained on data generated from GPT4, among other models.

Evals

Training

See axolotl config

axolotl version: 0.4.0

base_model: /workspace/axolotl/dbrx-checkpoint
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true

load_in_8bit: false
# load_in_4bit: true
strict: false

# adapter: qlora
# lora_modules_to_save: [embed_tokens, lm_head]

# lora_r: 32
# lora_alpha: 16
# lora_dropout: 0.05
# lora_target_linear: false
# lora_fan_in_fan_out:

datasets:
  - path: /workspace/datasets/dolphin-2.9/dolphin201-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  # - path: /workspace/datasets/dolphin-2.9/Ultrachat200kunfiltered.jsonl
  #   type: sharegpt
  #   conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/dolphin-coder-translate-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/dolphin-coder-codegen-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/m-a-p_Code-Feedback-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/m-a-p_CodeFeedback-Filtered-Instruction-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/not_samantha_norefusals.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/Orca-Math-resort-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/agent_instruct_react_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_instruct_j1s1_3k_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_negative_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_react_10p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_tflan_cot_30p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/openhermes200k_unfiltered.jsonl
    type: sharegpt 
    conversation: chatml
  # - path: /workspace/datasets/dolphin-2.9/SystemConversations.jsonl
  #   type: sharegpt
  #   conversation: chatml

chat_template: chatml

unfrozen_parameters:
- ^lm_head.weight$
# ffn.experts.mlp_experts.0.v1 layers
- transformer.blocks.30.ffn.experts.mlp_experts.0.v1
- transformer.blocks.32.ffn.experts.mlp_experts.0.v1
- transformer.blocks.25.ffn.experts.mlp_experts.0.v1
- transformer.blocks.15.ffn.experts.mlp_experts.0.v1
- transformer.blocks.22.ffn.experts.mlp_experts.0.v1
- transformer.blocks.31.ffn.experts.mlp_experts.0.v1
- transformer.blocks.7.ffn.experts.mlp_experts.0.v1
- transformer.blocks.21.ffn.experts.mlp_experts.0.v1
- transformer.blocks.8.ffn.experts.mlp_experts.0.v1
- transformer.blocks.23.ffn.experts.mlp_experts.0.v1
# ffn.experts.mlp_experts.0.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.0.w1
- transformer.blocks.8.ffn.experts.mlp_experts.0.w1
- transformer.blocks.30.ffn.experts.mlp_experts.0.w1
- transformer.blocks.4.ffn.experts.mlp_experts.0.w1
- transformer.blocks.0.ffn.experts.mlp_experts.0.w1
- transformer.blocks.32.ffn.experts.mlp_experts.0.w1
- transformer.blocks.6.ffn.experts.mlp_experts.0.w1
- transformer.blocks.3.ffn.experts.mlp_experts.0.w1
- transformer.blocks.25.ffn.experts.mlp_experts.0.w1
- transformer.blocks.5.ffn.experts.mlp_experts.0.w1
# ffn.experts.mlp_experts.0.w2 layers
- transformer.blocks.25.ffn.experts.mlp_experts.0.w2
- transformer.blocks.22.ffn.experts.mlp_experts.0.w2
- transformer.blocks.27.ffn.experts.mlp_experts.0.w2
- transformer.blocks.26.ffn.experts.mlp_experts.0.w2
- transformer.blocks.4.ffn.experts.mlp_experts.0.w2
- transformer.blocks.29.ffn.experts.mlp_experts.0.w2
- transformer.blocks.32.ffn.experts.mlp_experts.0.w2
- transformer.blocks.5.ffn.experts.mlp_experts.0.w2
- transformer.blocks.7.ffn.experts.mlp_experts.0.w2
- transformer.blocks.3.ffn.experts.mlp_experts.0.w2
# ffn.experts.mlp_experts.1.v1 layers
- transformer.blocks.27.ffn.experts.mlp_experts.1.v1
- transformer.blocks.25.ffn.experts.mlp_experts.1.v1
- transformer.blocks.29.ffn.experts.mlp_experts.1.v1
- transformer.blocks.33.ffn.experts.mlp_experts.1.v1
- transformer.blocks.23.ffn.experts.mlp_experts.1.v1
- transformer.blocks.30.ffn.experts.mlp_experts.1.v1
- transformer.blocks.6.ffn.experts.mlp_experts.1.v1
- transformer.blocks.21.ffn.experts.mlp_experts.1.v1
- transformer.blocks.15.ffn.experts.mlp_experts.1.v1
- transformer.blocks.7.ffn.experts.mlp_experts.1.v1
# ffn.experts.mlp_experts.1.w1 layers
- transformer.blocks.0.ffn.experts.mlp_experts.1.w1
- transformer.blocks.6.ffn.experts.mlp_experts.1.w1
- transformer.blocks.7.ffn.experts.mlp_experts.1.w1
- transformer.blocks.4.ffn.experts.mlp_experts.1.w1
- transformer.blocks.8.ffn.experts.mlp_experts.1.w1
- transformer.blocks.29.ffn.experts.mlp_experts.1.w1
- transformer.blocks.33.ffn.experts.mlp_experts.1.w1
- transformer.blocks.27.ffn.experts.mlp_experts.1.w1
- transformer.blocks.1.ffn.experts.mlp_experts.1.w1
- transformer.blocks.10.ffn.experts.mlp_experts.1.w1
# ffn.experts.mlp_experts.1.w2 layers
- transformer.blocks.25.ffn.experts.mlp_experts.1.w2
- transformer.blocks.23.ffn.experts.mlp_experts.1.w2
- transformer.blocks.27.ffn.experts.mlp_experts.1.w2
- transformer.blocks.29.ffn.experts.mlp_experts.1.w2
- transformer.blocks.31.ffn.experts.mlp_experts.1.w2
- transformer.blocks.4.ffn.experts.mlp_experts.1.w2
- transformer.blocks.32.ffn.experts.mlp_experts.1.w2
- transformer.blocks.30.ffn.experts.mlp_experts.1.w2
- transformer.blocks.21.ffn.experts.mlp_experts.1.w2
- transformer.blocks.33.ffn.experts.mlp_experts.1.w2
# ffn.experts.mlp_experts.10.v1 layers
- transformer.blocks.28.ffn.experts.mlp_experts.10.v1
- transformer.blocks.34.ffn.experts.mlp_experts.10.v1
- transformer.blocks.33.ffn.experts.mlp_experts.10.v1
- transformer.blocks.26.ffn.experts.mlp_experts.10.v1
- transformer.blocks.32.ffn.experts.mlp_experts.10.v1
- transformer.blocks.30.ffn.experts.mlp_experts.10.v1
- transformer.blocks.36.ffn.experts.mlp_experts.10.v1
- transformer.blocks.24.ffn.experts.mlp_experts.10.v1
- transformer.blocks.20.ffn.experts.mlp_experts.10.v1
- transformer.blocks.35.ffn.experts.mlp_experts.10.v1
# ffn.experts.mlp_experts.10.w1 layers
- transformer.blocks.24.ffn.experts.mlp_experts.10.w1
- transformer.blocks.33.ffn.experts.mlp_experts.10.w1
- transformer.blocks.8.ffn.experts.mlp_experts.10.w1
- transformer.blocks.7.ffn.experts.mlp_experts.10.w1
- transformer.blocks.34.ffn.experts.mlp_experts.10.w1
- transformer.blocks.28.ffn.experts.mlp_experts.10.w1
- transformer.blocks.30.ffn.experts.mlp_experts.10.w1
- transformer.blocks.1.ffn.experts.mlp_experts.10.w1
- transformer.blocks.3.ffn.experts.mlp_experts.10.w1
- transformer.blocks.5.ffn.experts.mlp_experts.10.w1
# ffn.experts.mlp_experts.10.w2 layers
- transformer.blocks.24.ffn.experts.mlp_experts.10.w2
- transformer.blocks.28.ffn.experts.mlp_experts.10.w2
- transformer.blocks.23.ffn.experts.mlp_experts.10.w2
- transformer.blocks.30.ffn.experts.mlp_experts.10.w2
- transformer.blocks.32.ffn.experts.mlp_experts.10.w2
- transformer.blocks.3.ffn.experts.mlp_experts.10.w2
- transformer.blocks.33.ffn.experts.mlp_experts.10.w2
- transformer.blocks.26.ffn.experts.mlp_experts.10.w2
- transformer.blocks.2.ffn.experts.mlp_experts.10.w2
- transformer.blocks.20.ffn.experts.mlp_experts.10.w2
# ffn.experts.mlp_experts.11.w1 layers
- transformer.blocks.6.ffn.experts.mlp_experts.11.w1
- transformer.blocks.8.ffn.experts.mlp_experts.11.w1
- transformer.blocks.9.ffn.experts.mlp_experts.11.w1
- transformer.blocks.0.ffn.experts.mlp_experts.11.w1
- transformer.blocks.10.ffn.experts.mlp_experts.11.w1
- transformer.blocks.28.ffn.experts.mlp_experts.11.w1
- transformer.blocks.3.ffn.experts.mlp_experts.11.w1
- transformer.blocks.5.ffn.experts.mlp_experts.11.w1
- transformer.blocks.33.ffn.experts.mlp_experts.11.w1
- transformer.blocks.13.ffn.experts.mlp_experts.11.w1
# ffn.experts.mlp_experts.11.w2 layers
- transformer.blocks.27.ffn.experts.mlp_experts.11.w2
- transformer.blocks.24.ffn.experts.mlp_experts.11.w2
- transformer.blocks.29.ffn.experts.mlp_experts.11.w2
- transformer.blocks.30.ffn.experts.mlp_experts.11.w2
- transformer.blocks.22.ffn.experts.mlp_experts.11.w2
- transformer.blocks.6.ffn.experts.mlp_experts.11.w2
- transformer.blocks.25.ffn.experts.mlp_experts.11.w2
- transformer.blocks.7.ffn.experts.mlp_experts.11.w2
- transformer.blocks.28.ffn.experts.mlp_experts.11.w2
- transformer.blocks.5.ffn.experts.mlp_experts.11.w2
# ffn.experts.mlp_experts.12.v1 layers
- transformer.blocks.30.ffn.experts.mlp_experts.12.v1
- transformer.blocks.21.ffn.experts.mlp_experts.12.v1
- transformer.blocks.27.ffn.experts.mlp_experts.12.v1
- transformer.blocks.28.ffn.experts.mlp_experts.12.v1
- transformer.blocks.29.ffn.experts.mlp_experts.12.v1
- transformer.blocks.8.ffn.experts.mlp_experts.12.v1
- transformer.blocks.10.ffn.experts.mlp_experts.12.v1
- transformer.blocks.23.ffn.experts.mlp_experts.12.v1
- transformer.blocks.6.ffn.experts.mlp_experts.12.v1
- transformer.blocks.20.ffn.experts.mlp_experts.12.v1
# ffn.experts.mlp_experts.12.w1 layers
- transformer.blocks.8.ffn.experts.mlp_experts.12.w1
- transformer.blocks.1.ffn.experts.mlp_experts.12.w1
- transformer.blocks.0.ffn.experts.mlp_experts.12.w1
- transformer.blocks.6.ffn.experts.mlp_experts.12.w1
- transformer.blocks.9.ffn.experts.mlp_experts.12.w1
- transformer.blocks.2.ffn.experts.mlp_experts.12.w1
- transformer.blocks.10.ffn.experts.mlp_experts.12.w1
- transformer.blocks.17.ffn.experts.mlp_experts.12.w1
- transformer.blocks.29.ffn.experts.mlp_experts.12.w1
- transformer.blocks.21.ffn.experts.mlp_experts.12.w1
# ffn.experts.mlp_experts.12.w2 layers
- transformer.blocks.6.ffn.experts.mlp_experts.12.w2
- transformer.blocks.25.ffn.experts.mlp_experts.12.w2
- transformer.blocks.27.ffn.experts.mlp_experts.12.w2
- transformer.blocks.8.ffn.experts.mlp_experts.12.w2
- transformer.blocks.31.ffn.experts.mlp_experts.12.w2
- transformer.blocks.21.ffn.experts.mlp_experts.12.w2
- transformer.blocks.2.ffn.experts.mlp_experts.12.w2
- transformer.blocks.29.ffn.experts.mlp_experts.12.w2
- transformer.blocks.32.ffn.experts.mlp_experts.12.w2
- transformer.blocks.30.ffn.experts.mlp_experts.12.w2
# ffn.experts.mlp_experts.13.v1 layers
- transformer.blocks.31.ffn.experts.mlp_experts.13.v1
- transformer.blocks.24.ffn.experts.mlp_experts.13.v1
- transformer.blocks.30.ffn.experts.mlp_experts.13.v1
- transformer.blocks.29.ffn.experts.mlp_experts.13.v1
- transformer.blocks.8.ffn.experts.mlp_experts.13.v1
- transformer.blocks.10.ffn.experts.mlp_experts.13.v1
- transformer.blocks.11.ffn.experts.mlp_experts.13.v1
- transformer.blocks.27.ffn.experts.mlp_experts.13.v1
- transformer.blocks.25.ffn.experts.mlp_experts.13.v1
- transformer.blocks.36.ffn.experts.mlp_experts.13.v1
# ffn.experts.mlp_experts.13.w1 layers
- transformer.blocks.4.ffn.experts.mlp_experts.13.w1
- transformer.blocks.10.ffn.experts.mlp_experts.13.w1
- transformer.blocks.6.ffn.experts.mlp_experts.13.w1
- transformer.blocks.0.ffn.experts.mlp_experts.13.w1
- transformer.blocks.3.ffn.experts.mlp_experts.13.w1
- transformer.blocks.24.ffn.experts.mlp_experts.13.w1
- transformer.blocks.8.ffn.experts.mlp_experts.13.w1
- transformer.blocks.1.ffn.experts.mlp_experts.13.w1
- transformer.blocks.30.ffn.experts.mlp_experts.13.w1
- transformer.blocks.11.ffn.experts.mlp_experts.13.w1
# ffn.experts.mlp_experts.13.w2 layers
- transformer.blocks.24.ffn.experts.mlp_experts.13.w2
- transformer.blocks.20.ffn.experts.mlp_experts.13.w2
- transformer.blocks.25.ffn.experts.mlp_experts.13.w2
- transformer.blocks.27.ffn.experts.mlp_experts.13.w2
- transformer.blocks.3.ffn.experts.mlp_experts.13.w2
- transformer.blocks.4.ffn.experts.mlp_experts.13.w2
- transformer.blocks.29.ffn.experts.mlp_experts.13.w2
- transformer.blocks.6.ffn.experts.mlp_experts.13.w2
- transformer.blocks.30.ffn.experts.mlp_experts.13.w2
- transformer.blocks.31.ffn.experts.mlp_experts.13.w2
# ffn.experts.mlp_experts.14.v1 layers
- transformer.blocks.28.ffn.experts.mlp_experts.14.v1
- transformer.blocks.26.ffn.experts.mlp_experts.14.v1
- transformer.blocks.29.ffn.experts.mlp_experts.14.v1
- transformer.blocks.35.ffn.experts.mlp_experts.14.v1
- transformer.blocks.24.ffn.experts.mlp_experts.14.v1
- transformer.blocks.8.ffn.experts.mlp_experts.14.v1
- transformer.blocks.32.ffn.experts.mlp_experts.14.v1
- transformer.blocks.15.ffn.experts.mlp_experts.14.v1
- transformer.blocks.11.ffn.experts.mlp_experts.14.v1
- transformer.blocks.22.ffn.experts.mlp_experts.14.v1
# ffn.experts.mlp_experts.14.w1 layers
- transformer.blocks.8.ffn.experts.mlp_experts.14.w1
- transformer.blocks.4.ffn.experts.mlp_experts.14.w1
- transformer.blocks.5.ffn.experts.mlp_experts.14.w1
- transformer.blocks.7.ffn.experts.mlp_experts.14.w1
- transformer.blocks.3.ffn.experts.mlp_experts.14.w1
- transformer.blocks.13.ffn.experts.mlp_experts.14.w1
- transformer.blocks.29.ffn.experts.mlp_experts.14.w1
- transformer.blocks.6.ffn.experts.mlp_experts.14.w1
- transformer.blocks.28.ffn.experts.mlp_experts.14.w1
- transformer.blocks.9.ffn.experts.mlp_experts.14.w1
# ffn.experts.mlp_experts.14.w2 layers
- transformer.blocks.26.ffn.experts.mlp_experts.14.w2
- transformer.blocks.24.ffn.experts.mlp_experts.14.w2
- transformer.blocks.29.ffn.experts.mlp_experts.14.w2
- transformer.blocks.28.ffn.experts.mlp_experts.14.w2
- transformer.blocks.31.ffn.experts.mlp_experts.14.w2
- transformer.blocks.5.ffn.experts.mlp_experts.14.w2
- transformer.blocks.4.ffn.experts.mlp_experts.14.w2
- transformer.blocks.32.ffn.experts.mlp_experts.14.w2
- transformer.blocks.6.ffn.experts.mlp_experts.14.w2
- transformer.blocks.22.ffn.experts.mlp_experts.14.w2
# ffn.experts.mlp_experts.15.v1 layers
- transformer.blocks.33.ffn.experts.mlp_experts.15.v1
- transformer.blocks.26.ffn.experts.mlp_experts.15.v1
- transformer.blocks.31.ffn.experts.mlp_experts.15.v1
- transformer.blocks.28.ffn.experts.mlp_experts.15.v1
- transformer.blocks.9.ffn.experts.mlp_experts.15.v1
- transformer.blocks.34.ffn.experts.mlp_experts.15.v1
- transformer.blocks.29.ffn.experts.mlp_experts.15.v1
- transformer.blocks.7.ffn.experts.mlp_experts.15.v1
- transformer.blocks.17.ffn.experts.mlp_experts.15.v1
- transformer.blocks.15.ffn.experts.mlp_experts.15.v1
# ffn.experts.mlp_experts.15.w1 layers
- transformer.blocks.6.ffn.experts.mlp_experts.15.w1
- transformer.blocks.9.ffn.experts.mlp_experts.15.w1
- transformer.blocks.0.ffn.experts.mlp_experts.15.w1
- transformer.blocks.7.ffn.experts.mlp_experts.15.w1
- transformer.blocks.14.ffn.experts.mlp_experts.15.w1
- transformer.blocks.33.ffn.experts.mlp_experts.15.w1
- transformer.blocks.34.ffn.experts.mlp_experts.15.w1
- transformer.blocks.10.ffn.experts.mlp_experts.15.w1
- transformer.blocks.5.ffn.experts.mlp_experts.15.w1
- transformer.blocks.29.ffn.experts.mlp_experts.15.w1
# ffn.experts.mlp_experts.15.w2 layers
- transformer.blocks.28.ffn.experts.mlp_experts.15.w2
- transformer.blocks.26.ffn.experts.mlp_experts.15.w2
- transformer.blocks.27.ffn.experts.mlp_experts.15.w2
- transformer.blocks.29.ffn.experts.mlp_experts.15.w2
- transformer.blocks.6.ffn.experts.mlp_experts.15.w2
- transformer.blocks.31.ffn.experts.mlp_experts.15.w2
- transformer.blocks.7.ffn.experts.mlp_experts.15.w2
- transformer.blocks.33.ffn.experts.mlp_experts.15.w2
- transformer.blocks.32.ffn.experts.mlp_experts.15.w2
- transformer.blocks.25.ffn.experts.mlp_experts.15.w2
# ffn.experts.mlp_experts.2.v1 layers
- transformer.blocks.31.ffn.experts.mlp_experts.2.v1
- transformer.blocks.27.ffn.experts.mlp_experts.2.v1
- transformer.blocks.28.ffn.experts.mlp_experts.2.v1
- transformer.blocks.30.ffn.experts.mlp_experts.2.v1
- transformer.blocks.23.ffn.experts.mlp_experts.2.v1
- transformer.blocks.32.ffn.experts.mlp_experts.2.v1
- transformer.blocks.35.ffn.experts.mlp_experts.2.v1
- transformer.blocks.7.ffn.experts.mlp_experts.2.v1
- transformer.blocks.21.ffn.experts.mlp_experts.2.v1
- transformer.blocks.15.ffn.experts.mlp_experts.2.v1
# ffn.experts.mlp_experts.2.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.2.w1
- transformer.blocks.6.ffn.experts.mlp_experts.2.w1
- transformer.blocks.1.ffn.experts.mlp_experts.2.w1
- transformer.blocks.4.ffn.experts.mlp_experts.2.w1
- transformer.blocks.5.ffn.experts.mlp_experts.2.w1
- transformer.blocks.29.ffn.experts.mlp_experts.2.w1
- transformer.blocks.0.ffn.experts.mlp_experts.2.w1
- transformer.blocks.9.ffn.experts.mlp_experts.2.w1
- transformer.blocks.31.ffn.experts.mlp_experts.2.w1
- transformer.blocks.30.ffn.experts.mlp_experts.2.w1
# ffn.experts.mlp_experts.2.w2 layers
- transformer.blocks.26.ffn.experts.mlp_experts.2.w2
- transformer.blocks.27.ffn.experts.mlp_experts.2.w2
- transformer.blocks.33.ffn.experts.mlp_experts.2.w2
- transformer.blocks.5.ffn.experts.mlp_experts.2.w2
- transformer.blocks.23.ffn.experts.mlp_experts.2.w2
- transformer.blocks.32.ffn.experts.mlp_experts.2.w2
- transformer.blocks.28.ffn.experts.mlp_experts.2.w2
- transformer.blocks.4.ffn.experts.mlp_experts.2.w2
- transformer.blocks.29.ffn.experts.mlp_experts.2.w2
- transformer.blocks.30.ffn.experts.mlp_experts.2.w2
# ffn.experts.mlp_experts.3.v1 layers
- transformer.blocks.28.ffn.experts.mlp_experts.3.v1
- transformer.blocks.33.ffn.experts.mlp_experts.3.v1
- transformer.blocks.36.ffn.experts.mlp_experts.3.v1
- transformer.blocks.29.ffn.experts.mlp_experts.3.v1
- transformer.blocks.30.ffn.experts.mlp_experts.3.v1
- transformer.blocks.7.ffn.experts.mlp_experts.3.v1
- transformer.blocks.14.ffn.experts.mlp_experts.3.v1
- transformer.blocks.10.ffn.experts.mlp_experts.3.v1
- transformer.blocks.31.ffn.experts.mlp_experts.3.v1
- transformer.blocks.21.ffn.experts.mlp_experts.3.v1
# ffn.experts.mlp_experts.3.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.3.w1
- transformer.blocks.0.ffn.experts.mlp_experts.3.w1
- transformer.blocks.10.ffn.experts.mlp_experts.3.w1
- transformer.blocks.9.ffn.experts.mlp_experts.3.w1
- transformer.blocks.29.ffn.experts.mlp_experts.3.w1
- transformer.blocks.5.ffn.experts.mlp_experts.3.w1
- transformer.blocks.30.ffn.experts.mlp_experts.3.w1
- transformer.blocks.4.ffn.experts.mlp_experts.3.w1
- transformer.blocks.33.ffn.experts.mlp_experts.3.w1
- transformer.blocks.1.ffn.experts.mlp_experts.3.w1
# ffn.experts.mlp_experts.3.w2 layers
- transformer.blocks.28.ffn.experts.mlp_experts.3.w2
- transformer.blocks.5.ffn.experts.mlp_experts.3.w2
- transformer.blocks.24.ffn.experts.mlp_experts.3.w2
- transformer.blocks.31.ffn.experts.mlp_experts.3.w2
- transformer.blocks.30.ffn.experts.mlp_experts.3.w2
- transformer.blocks.21.ffn.experts.mlp_experts.3.w2
- transformer.blocks.32.ffn.experts.mlp_experts.3.w2
- transformer.blocks.29.ffn.experts.mlp_experts.3.w2
- transformer.blocks.26.ffn.experts.mlp_experts.3.w2
- transformer.blocks.2.ffn.experts.mlp_experts.3.w2
# ffn.experts.mlp_experts.4.v1 layers
- transformer.blocks.34.ffn.experts.mlp_experts.4.v1
- transformer.blocks.31.ffn.experts.mlp_experts.4.v1
- transformer.blocks.26.ffn.experts.mlp_experts.4.v1
- transformer.blocks.24.ffn.experts.mlp_experts.4.v1
- transformer.blocks.14.ffn.experts.mlp_experts.4.v1
- transformer.blocks.32.ffn.experts.mlp_experts.4.v1
- transformer.blocks.7.ffn.experts.mlp_experts.4.v1
- transformer.blocks.6.ffn.experts.mlp_experts.4.v1
- transformer.blocks.20.ffn.experts.mlp_experts.4.v1
- transformer.blocks.9.ffn.experts.mlp_experts.4.v1
# ffn.experts.mlp_experts.4.w1 layers
- transformer.blocks.6.ffn.experts.mlp_experts.4.w1
- transformer.blocks.4.ffn.experts.mlp_experts.4.w1
- transformer.blocks.7.ffn.experts.mlp_experts.4.w1
- transformer.blocks.9.ffn.experts.mlp_experts.4.w1
- transformer.blocks.0.ffn.experts.mlp_experts.4.w1
- transformer.blocks.5.ffn.experts.mlp_experts.4.w1
- transformer.blocks.14.ffn.experts.mlp_experts.4.w1
- transformer.blocks.34.ffn.experts.mlp_experts.4.w1
- transformer.blocks.8.ffn.experts.mlp_experts.4.w1
- transformer.blocks.29.ffn.experts.mlp_experts.4.w1
# ffn.experts.mlp_experts.4.w2 layers
- transformer.blocks.25.ffn.experts.mlp_experts.4.w2
- transformer.blocks.24.ffn.experts.mlp_experts.4.w2
- transformer.blocks.26.ffn.experts.mlp_experts.4.w2
- transformer.blocks.5.ffn.experts.mlp_experts.4.w2
- transformer.blocks.6.ffn.experts.mlp_experts.4.w2
- transformer.blocks.32.ffn.experts.mlp_experts.4.w2
- transformer.blocks.4.ffn.experts.mlp_experts.4.w2
- transformer.blocks.36.ffn.experts.mlp_experts.4.w2
- transformer.blocks.29.ffn.experts.mlp_experts.4.w2
- transformer.blocks.27.ffn.experts.mlp_experts.4.w2
# ffn.experts.mlp_experts.5.v1 layers
- transformer.blocks.35.ffn.experts.mlp_experts.5.v1
- transformer.blocks.30.ffn.experts.mlp_experts.5.v1
- transformer.blocks.28.ffn.experts.mlp_experts.5.v1
- transformer.blocks.32.ffn.experts.mlp_experts.5.v1
- transformer.blocks.27.ffn.experts.mlp_experts.5.v1
- transformer.blocks.26.ffn.experts.mlp_experts.5.v1
- transformer.blocks.33.ffn.experts.mlp_experts.5.v1
- transformer.blocks.29.ffn.experts.mlp_experts.5.v1
- transformer.blocks.8.ffn.experts.mlp_experts.5.v1
- transformer.blocks.7.ffn.experts.mlp_experts.5.v1
# ffn.experts.mlp_experts.5.w1 layers
- transformer.blocks.0.ffn.experts.mlp_experts.5.w1
- transformer.blocks.6.ffn.experts.mlp_experts.5.w1
- transformer.blocks.7.ffn.experts.mlp_experts.5.w1
- transformer.blocks.9.ffn.experts.mlp_experts.5.w1
- transformer.blocks.8.ffn.experts.mlp_experts.5.w1
- transformer.blocks.12.ffn.experts.mlp_experts.5.w1
- transformer.blocks.3.ffn.experts.mlp_experts.5.w1
- transformer.blocks.5.ffn.experts.mlp_experts.5.w1
- transformer.blocks.4.ffn.experts.mlp_experts.5.w1
- transformer.blocks.33.ffn.experts.mlp_experts.5.w1
# ffn.experts.mlp_experts.5.w2 layers
- transformer.blocks.26.ffn.experts.mlp_experts.5.w2
- transformer.blocks.28.ffn.experts.mlp_experts.5.w2
- transformer.blocks.6.ffn.experts.mlp_experts.5.w2
- transformer.blocks.33.ffn.experts.mlp_experts.5.w2
- transformer.blocks.5.ffn.experts.mlp_experts.5.w2
- transformer.blocks.27.ffn.experts.mlp_experts.5.w2
- transformer.blocks.3.ffn.experts.mlp_experts.5.w2
- transformer.blocks.29.ffn.experts.mlp_experts.5.w2
- transformer.blocks.25.ffn.experts.mlp_experts.5.w2
- transformer.blocks.7.ffn.experts.mlp_experts.5.w2
# ffn.experts.mlp_experts.6.v1 layers
- transformer.blocks.34.ffn.experts.mlp_experts.6.v1
- transformer.blocks.31.ffn.experts.mlp_experts.6.v1
- transformer.blocks.30.ffn.experts.mlp_experts.6.v1
- transformer.blocks.26.ffn.experts.mlp_experts.6.v1
- transformer.blocks.35.ffn.experts.mlp_experts.6.v1
- transformer.blocks.20.ffn.experts.mlp_experts.6.v1
- transformer.blocks.15.ffn.experts.mlp_experts.6.v1
- transformer.blocks.29.ffn.experts.mlp_experts.6.v1
- transformer.blocks.10.ffn.experts.mlp_experts.6.v1
- transformer.blocks.24.ffn.experts.mlp_experts.6.v1
# ffn.experts.mlp_experts.6.w1 layers
- transformer.blocks.0.ffn.experts.mlp_experts.6.w1
- transformer.blocks.10.ffn.experts.mlp_experts.6.w1
- transformer.blocks.9.ffn.experts.mlp_experts.6.w1
- transformer.blocks.30.ffn.experts.mlp_experts.6.w1
- transformer.blocks.4.ffn.experts.mlp_experts.6.w1
- transformer.blocks.34.ffn.experts.mlp_experts.6.w1
- transformer.blocks.26.ffn.experts.mlp_experts.6.w1
- transformer.blocks.2.ffn.experts.mlp_experts.6.w1
- transformer.blocks.29.ffn.experts.mlp_experts.6.w1
- transformer.blocks.8.ffn.experts.mlp_experts.6.w1
# ffn.experts.mlp_experts.6.w2 layers
- transformer.blocks.24.ffn.experts.mlp_experts.6.w2
- transformer.blocks.26.ffn.experts.mlp_experts.6.w2
- transformer.blocks.32.ffn.experts.mlp_experts.6.w2
- transformer.blocks.30.ffn.experts.mlp_experts.6.w2
- transformer.blocks.25.ffn.experts.mlp_experts.6.w2
- transformer.blocks.31.ffn.experts.mlp_experts.6.w2
- transformer.blocks.20.ffn.experts.mlp_experts.6.w2
- transformer.blocks.4.ffn.experts.mlp_experts.6.w2
- transformer.blocks.2.ffn.experts.mlp_experts.6.w2
- transformer.blocks.9.ffn.experts.mlp_experts.6.w2
# ffn.experts.mlp_experts.7.v1 layers
- transformer.blocks.27.ffn.experts.mlp_experts.7.v1
- transformer.blocks.28.ffn.experts.mlp_experts.7.v1
- transformer.blocks.33.ffn.experts.mlp_experts.7.v1
- transformer.blocks.29.ffn.experts.mlp_experts.7.v1
- transformer.blocks.24.ffn.experts.mlp_experts.7.v1
- transformer.blocks.11.ffn.experts.mlp_experts.7.v1
- transformer.blocks.12.ffn.experts.mlp_experts.7.v1
- transformer.blocks.10.ffn.experts.mlp_experts.7.v1
- transformer.blocks.23.ffn.experts.mlp_experts.7.v1
- transformer.blocks.34.ffn.experts.mlp_experts.7.v1
# ffn.experts.mlp_experts.7.w1 layers
- transformer.blocks.12.ffn.experts.mlp_experts.7.w1
- transformer.blocks.0.ffn.experts.mlp_experts.7.w1
- transformer.blocks.5.ffn.experts.mlp_experts.7.w1
- transformer.blocks.29.ffn.experts.mlp_experts.7.w1
- transformer.blocks.10.ffn.experts.mlp_experts.7.w1
- transformer.blocks.4.ffn.experts.mlp_experts.7.w1
- transformer.blocks.3.ffn.experts.mlp_experts.7.w1
- transformer.blocks.8.ffn.experts.mlp_experts.7.w1
- transformer.blocks.34.ffn.experts.mlp_experts.7.w1
- transformer.blocks.33.ffn.experts.mlp_experts.7.w1
# ffn.experts.mlp_experts.7.w2 layers
- transformer.blocks.23.ffn.experts.mlp_experts.7.w2
- transformer.blocks.24.ffn.experts.mlp_experts.7.w2
- transformer.blocks.31.ffn.experts.mlp_experts.7.w2
- transformer.blocks.28.ffn.experts.mlp_experts.7.w2
- transformer.blocks.27.ffn.experts.mlp_experts.7.w2
- transformer.blocks.5.ffn.experts.mlp_experts.7.w2
- transformer.blocks.25.ffn.experts.mlp_experts.7.w2
- transformer.blocks.29.ffn.experts.mlp_experts.7.w2
- transformer.blocks.3.ffn.experts.mlp_experts.7.w2
- transformer.blocks.33.ffn.experts.mlp_experts.7.w2
# ffn.experts.mlp_experts.8.v1 layers
- transformer.blocks.30.ffn.experts.mlp_experts.8.v1
- transformer.blocks.27.ffn.experts.mlp_experts.8.v1
- transformer.blocks.20.ffn.experts.mlp_experts.8.v1
- transformer.blocks.32.ffn.experts.mlp_experts.8.v1
- transformer.blocks.34.ffn.experts.mlp_experts.8.v1
- transformer.blocks.33.ffn.experts.mlp_experts.8.v1
- transformer.blocks.9.ffn.experts.mlp_experts.8.v1
- transformer.blocks.7.ffn.experts.mlp_experts.8.v1
- transformer.blocks.6.ffn.experts.mlp_experts.8.v1
- transformer.blocks.24.ffn.experts.mlp_experts.8.v1
# ffn.experts.mlp_experts.8.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.8.w1
- transformer.blocks.6.ffn.experts.mlp_experts.8.w1
- transformer.blocks.0.ffn.experts.mlp_experts.8.w1
- transformer.blocks.9.ffn.experts.mlp_experts.8.w1
- transformer.blocks.3.ffn.experts.mlp_experts.8.w1
- transformer.blocks.2.ffn.experts.mlp_experts.8.w1
- transformer.blocks.8.ffn.experts.mlp_experts.8.w1
- transformer.blocks.30.ffn.experts.mlp_experts.8.w1
- transformer.blocks.24.ffn.experts.mlp_experts.8.w1
- transformer.blocks.1.ffn.experts.mlp_experts.8.w1
# ffn.experts.mlp_experts.8.w2 layers
- transformer.blocks.32.ffn.experts.mlp_experts.8.w2
- transformer.blocks.24.ffn.experts.mlp_experts.8.w2
- transformer.blocks.27.ffn.experts.mlp_experts.8.w2
- transformer.blocks.30.ffn.experts.mlp_experts.8.w2
- transformer.blocks.31.ffn.experts.mlp_experts.8.w2
- transformer.blocks.28.ffn.experts.mlp_experts.8.w2
- transformer.blocks.2.ffn.experts.mlp_experts.8.w2
- transformer.blocks.3.ffn.experts.mlp_experts.8.w2
- transformer.blocks.23.ffn.experts.mlp_experts.8.w2
- transformer.blocks.29.ffn.experts.mlp_experts.8.w2
# ffn.experts.mlp_experts.9.v1 layers
- transformer.blocks.31.ffn.experts.mlp_experts.9.v1
- transformer.blocks.27.ffn.experts.mlp_experts.9.v1
- transformer.blocks.29.ffn.experts.mlp_experts.9.v1
- transformer.blocks.33.ffn.experts.mlp_experts.9.v1
- transformer.blocks.25.ffn.experts.mlp_experts.9.v1
- transformer.blocks.14.ffn.experts.mlp_experts.9.v1
- transformer.blocks.32.ffn.experts.mlp_experts.9.v1
- transformer.blocks.7.ffn.experts.mlp_experts.9.v1
- transformer.blocks.9.ffn.experts.mlp_experts.9.v1
- transformer.blocks.34.ffn.experts.mlp_experts.9.v1
# ffn.experts.mlp_experts.9.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.9.w1
- transformer.blocks.1.ffn.experts.mlp_experts.9.w1
- transformer.blocks.9.ffn.experts.mlp_experts.9.w1
- transformer.blocks.2.ffn.experts.mlp_experts.9.w1
- transformer.blocks.27.ffn.experts.mlp_experts.9.w1
- transformer.blocks.12.ffn.experts.mlp_experts.9.w1
- transformer.blocks.4.ffn.experts.mlp_experts.9.w1
- transformer.blocks.6.ffn.experts.mlp_experts.9.w1
- transformer.blocks.19.ffn.experts.mlp_experts.9.w1
- transformer.blocks.8.ffn.experts.mlp_experts.9.w1
# ffn.experts.mlp_experts.9.w2 layers
- transformer.blocks.26.ffn.experts.mlp_experts.9.w2
- transformer.blocks.25.ffn.experts.mlp_experts.9.w2
- transformer.blocks.28.ffn.experts.mlp_experts.9.w2
- transformer.blocks.27.ffn.experts.mlp_experts.9.w2
- transformer.blocks.31.ffn.experts.mlp_experts.9.w2
- transformer.blocks.29.ffn.experts.mlp_experts.9.w2
- transformer.blocks.7.ffn.experts.mlp_experts.9.w2
- transformer.blocks.34.ffn.experts.mlp_experts.9.w2
- transformer.blocks.2.ffn.experts.mlp_experts.9.w2
- transformer.blocks.33.ffn.experts.mlp_experts.9.w2
# ffn.router.layer layers
- transformer.blocks.2.ffn.router.layer
- transformer.blocks.3.ffn.router.layer
- transformer.blocks.4.ffn.router.layer
- transformer.blocks.5.ffn.router.layer
- transformer.blocks.6.ffn.router.layer
- transformer.blocks.7.ffn.router.layer
- transformer.blocks.8.ffn.router.layer
- transformer.blocks.9.ffn.router.layer
- transformer.blocks.10.ffn.router.layer
- transformer.blocks.11.ffn.router.layer
# norm_attn_norm.attn.Wqkv layers
- transformer.blocks.16.norm_attn_norm.attn.Wqkv
- transformer.blocks.15.norm_attn_norm.attn.Wqkv
- transformer.blocks.11.norm_attn_norm.attn.Wqkv
- transformer.blocks.14.norm_attn_norm.attn.Wqkv
- transformer.blocks.12.norm_attn_norm.attn.Wqkv
- transformer.blocks.20.norm_attn_norm.attn.Wqkv
- transformer.blocks.10.norm_attn_norm.attn.Wqkv
- transformer.blocks.9.norm_attn_norm.attn.Wqkv
- transformer.blocks.19.norm_attn_norm.attn.Wqkv
- transformer.blocks.18.norm_attn_norm.attn.Wqkv
# norm_attn_norm.attn.out_proj layers
- transformer.blocks.1.norm_attn_norm.attn.out_proj
- transformer.blocks.18.norm_attn_norm.attn.out_proj
- transformer.blocks.2.norm_attn_norm.attn.out_proj
- transformer.blocks.16.norm_attn_norm.attn.out_proj
- transformer.blocks.0.norm_attn_norm.attn.out_proj
- transformer.blocks.39.norm_attn_norm.attn.out_proj
- transformer.blocks.23.norm_attn_norm.attn.out_proj
- transformer.blocks.8.norm_attn_norm.attn.out_proj
- transformer.blocks.24.norm_attn_norm.attn.out_proj
- transformer.blocks.19.norm_attn_norm.attn.out_proj
# norm_attn_norm.norm_1 layers
- transformer.blocks.0.norm_attn_norm.norm_1
- transformer.blocks.1.norm_attn_norm.norm_1
- transformer.blocks.2.norm_attn_norm.norm_1
- transformer.blocks.3.norm_attn_norm.norm_1
- transformer.blocks.4.norm_attn_norm.norm_1
- transformer.blocks.5.norm_attn_norm.norm_1
- transformer.blocks.6.norm_attn_norm.norm_1
- transformer.blocks.7.norm_attn_norm.norm_1
- transformer.blocks.8.norm_attn_norm.norm_1
- transformer.blocks.9.norm_attn_norm.norm_1
# norm_attn_norm.norm_2 layers
- transformer.blocks.0.norm_attn_norm.norm_2
- transformer.blocks.1.norm_attn_norm.norm_2
- transformer.blocks.2.norm_attn_norm.norm_2
- transformer.blocks.3.norm_attn_norm.norm_2
- transformer.blocks.4.norm_attn_norm.norm_2
- transformer.blocks.5.norm_attn_norm.norm_2
- transformer.blocks.6.norm_attn_norm.norm_2
- transformer.blocks.7.norm_attn_norm.norm_2
- transformer.blocks.8.norm_attn_norm.norm_2
- transformer.blocks.9.norm_attn_norm.norm_2
# transformer.norm_f layers
# transformer.wte layers
# ffn.experts.mlp_experts.11.v1 layers
- transformer.blocks.29.ffn.experts.mlp_experts.11.v1
- transformer.blocks.27.ffn.experts.mlp_experts.11.v1
- transformer.blocks.30.ffn.experts.mlp_experts.11.v1
- transformer.blocks.28.ffn.experts.mlp_experts.11.v1
- transformer.blocks.22.ffn.experts.mlp_experts.11.v1
- transformer.blocks.7.ffn.experts.mlp_experts.11.v1
- transformer.blocks.24.ffn.experts.mlp_experts.11.v1
- transformer.blocks.8.ffn.experts.mlp_experts.11.v1
- transformer.blocks.6.ffn.experts.mlp_experts.11.v1
- transformer.blocks.12.ffn.experts.mlp_experts.11.v1



dataset_prepared_path: dbrx2
val_set_size: 0.01
output_dir: ./out

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

wandb_project: dolphin-2.9-Dbrx
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 1
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 1e-5

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
# resume_from_checkpoint: /workspace/axolotl/dbrx-checkpoint
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 4
save_total_limit: 2
save_steps:
debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16_cpuoffload_params.json
weight_decay: 0.05
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<|endoftext|>"
  eos_token: "<|im_end|>"
  pad_token: "<|pad|>"
  unk_token: "<|endoftext|>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"

out

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.4336

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: 1e-05
train_batch_size: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 8
total_train_batch_size: 64
total_eval_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 10
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
0.4009	0.0	1	0.4328
0.413	0.25	587	0.4408
0.3626	0.5	1174	0.4368
0.3896	0.75	1761	0.4336

Framework versions

Transformers 4.40.0.dev0
Pytorch 2.2.2+cu121
Datasets 2.15.0
Tokenizers 0.15.0

Downloads last month: 18

Safetensors

Model size

132B params

Tensor type

BF16

Model tree for dphn/dolphin-2.9.1-dbrx

Base model

databricks/dbrx-base

Finetuned

(2)

this model

dphn
/

dolphin-2.9.1-dbrx

Dolphin 2.9.1 DBRX 🐬

Evals

Training

out

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Model tree for dphn/dolphin-2.9.1-dbrx

Datasets used to train dphn/dolphin-2.9.1-dbrx