Upload DASH-Q quantized checkpoint (INTNone, gNone)
Browse files- README.md +73 -0
- chat_template.jinja +1 -0
- config.json +56 -0
- dashq_config.json +1960 -0
- dashq_kernel.py +301 -0
- generation_config.json +11 -0
- model-00001-of-00002.safetensors +3 -0
- model-00002-of-00002.safetensors +3 -0
- model.safetensors.index.json +571 -0
- modeling_dashq.py +227 -0
- tokenizer.json +0 -0
- tokenizer_config.json +12 -0
README.md
ADDED
|
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: microsoft/phi-4
|
| 4 |
+
library_name: transformers
|
| 5 |
+
tags:
|
| 6 |
+
- dashq
|
| 7 |
+
- quantized
|
| 8 |
+
- post-training-quantization
|
| 9 |
+
---
|
| 10 |
+
# phi-4-DASHQ-INT4-g128
|
| 11 |
+
|
| 12 |
+
This repository contains a DASH-Q packed quantized checkpoint for `microsoft/phi-4`.
|
| 13 |
+
|
| 14 |
+
DASH-Q checkpoints require the lightweight DASH-Q runtime package for loading. They are not plain Transformers checkpoints because linear layers are stored as `PackedQuantizedLinear` modules.
|
| 15 |
+
|
| 16 |
+
## Usage
|
| 17 |
+
|
| 18 |
+
```python
|
| 19 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 20 |
+
|
| 21 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 22 |
+
"jkim96/phi-4-DASHQ-INT4-g128", trust_remote_code=True, device_map="cuda", dtype="auto"
|
| 23 |
+
)
|
| 24 |
+
tokenizer = AutoTokenizer.from_pretrained("jkim96/phi-4-DASHQ-INT4-g128")
|
| 25 |
+
|
| 26 |
+
messages = [{"role": "user", "content": "Explain 2-bit quantization in one sentence."}]
|
| 27 |
+
text = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
|
| 28 |
+
inputs = tokenizer(text, return_tensors="pt").to(model.device)
|
| 29 |
+
print(tokenizer.decode(model.generate(**inputs, max_new_tokens=256)[0]))
|
| 30 |
+
```
|
| 31 |
+
|
| 32 |
+
`trust_remote_code=True` is required: the checkpoint ships its quantized-layer
|
| 33 |
+
implementation (`modeling_dashq.py`) and Triton kernels (`dashq_kernel.py`).
|
| 34 |
+
Without Triton, or on CPU, it falls back to dequantize-and-matmul in PyTorch.
|
| 35 |
+
|
| 36 |
+
### Requirements
|
| 37 |
+
|
| 38 |
+
| Package | Minimum | Verified with |
|
| 39 |
+
| --- | --- | --- |
|
| 40 |
+
| `torch` | 2.4 | 2.12.1+cu130 |
|
| 41 |
+
| `transformers` | 5.8 | 5.9.0 |
|
| 42 |
+
| `triton` | 3.0 (Linux; bundled with CUDA builds of PyTorch) | 3.7.1 |
|
| 43 |
+
| `huggingface_hub` | 1.5 (pulled in by transformers) | 1.15.0 |
|
| 44 |
+
|
| 45 |
+
## Quantization
|
| 46 |
+
|
| 47 |
+
| Field | Value |
|
| 48 |
+
| --- | --- |
|
| 49 |
+
| Base model | `microsoft/phi-4` |
|
| 50 |
+
| Bits | `4` |
|
| 51 |
+
| Group size | `128` |
|
| 52 |
+
| Scale/zero dtype | `float16` |
|
| 53 |
+
| Calibration dataset | `wikitext2` |
|
| 54 |
+
| Calibration samples | `128` |
|
| 55 |
+
| Sequence length | `2048` |
|
| 56 |
+
| Original size | `29.3190 GB` |
|
| 57 |
+
| Quantized size | `9.2978 GB` |
|
| 58 |
+
|
| 59 |
+
## Evaluation
|
| 60 |
+
|
| 61 |
+
| Metric | Value |
|
| 62 |
+
| --- | ---: |
|
| 63 |
+
| `wikitext2_ppl` | 6.5151 |
|
| 64 |
+
| `zero-shot accuracy avg` | 69.0414 |
|
| 65 |
+
| `arc_challenge` | 56.2287 |
|
| 66 |
+
| `arc_easy` | 74.3266 |
|
| 67 |
+
| `commonsense_qa` | 74.4472 |
|
| 68 |
+
| `hellaswag` | 81.8462 |
|
| 69 |
+
| `lambada_openai` | 72.2880 |
|
| 70 |
+
| `openbookqa` | 45.4000 |
|
| 71 |
+
| `piqa` | 81.3384 |
|
| 72 |
+
| `truthfulqa_mc2` | 59.3332 |
|
| 73 |
+
| `winogrande` | 76.1642 |
|
chat_template.jinja
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{% for message in messages %}{% if (message['role'] == 'system') %}{{'<|im_start|>system<|im_sep|>' + message['content'] + '<|im_end|>'}}{% elif (message['role'] == 'user') %}{{'<|im_start|>user<|im_sep|>' + message['content'] + '<|im_end|>'}}{% elif (message['role'] == 'assistant') %}{{'<|im_start|>assistant<|im_sep|>' + message['content'] + '<|im_end|>'}}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant<|im_sep|>' }}{% endif %}
|
config.json
ADDED
|
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"DashQPhi3ForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": false,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"auto_map": {
|
| 8 |
+
"AutoModelForCausalLM": "modeling_dashq.DashQPhi3ForCausalLM"
|
| 9 |
+
},
|
| 10 |
+
"bos_token_id": 100257,
|
| 11 |
+
"dashq": {
|
| 12 |
+
"format": "dashq-packed-linear",
|
| 13 |
+
"format_version": 1,
|
| 14 |
+
"layer_metadata": "dashq_config.json",
|
| 15 |
+
"method": "dashq",
|
| 16 |
+
"n_quantized_modules": 160,
|
| 17 |
+
"params": {
|
| 18 |
+
"bits": 4,
|
| 19 |
+
"group_size": 128,
|
| 20 |
+
"low_memory_optimization": false,
|
| 21 |
+
"moe_hessian_scope": "shared",
|
| 22 |
+
"n_samples": 128,
|
| 23 |
+
"scale_zero_dtype": "float16",
|
| 24 |
+
"symmetric": false,
|
| 25 |
+
"use_error_compensation": true,
|
| 26 |
+
"use_optimal_shrinkage": true,
|
| 27 |
+
"use_weighted_quantization": true
|
| 28 |
+
}
|
| 29 |
+
},
|
| 30 |
+
"dtype": "bfloat16",
|
| 31 |
+
"embd_pdrop": 0.0,
|
| 32 |
+
"eos_token_id": 100265,
|
| 33 |
+
"hidden_act": "silu",
|
| 34 |
+
"hidden_size": 5120,
|
| 35 |
+
"initializer_range": 0.02,
|
| 36 |
+
"intermediate_size": 17920,
|
| 37 |
+
"max_position_embeddings": 16384,
|
| 38 |
+
"model_type": "phi3",
|
| 39 |
+
"num_attention_heads": 40,
|
| 40 |
+
"num_hidden_layers": 40,
|
| 41 |
+
"num_key_value_heads": 10,
|
| 42 |
+
"original_max_position_embeddings": 16384,
|
| 43 |
+
"pad_token_id": 100349,
|
| 44 |
+
"resid_pdrop": 0.0,
|
| 45 |
+
"rms_norm_eps": 1e-05,
|
| 46 |
+
"rope_parameters": {
|
| 47 |
+
"partial_rotary_factor": 1.0,
|
| 48 |
+
"rope_theta": 250000,
|
| 49 |
+
"rope_type": "default"
|
| 50 |
+
},
|
| 51 |
+
"sliding_window": null,
|
| 52 |
+
"tie_word_embeddings": false,
|
| 53 |
+
"transformers_version": "5.15.0",
|
| 54 |
+
"use_cache": false,
|
| 55 |
+
"vocab_size": 100352
|
| 56 |
+
}
|
dashq_config.json
ADDED
|
@@ -0,0 +1,1960 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"base_model": "microsoft/phi-4",
|
| 3 |
+
"format": "dashq-packed-linear",
|
| 4 |
+
"format_version": 1,
|
| 5 |
+
"method": "dashq",
|
| 6 |
+
"model_class": "causal_lm",
|
| 7 |
+
"params": {
|
| 8 |
+
"bits": 4,
|
| 9 |
+
"group_size": 128,
|
| 10 |
+
"low_memory_optimization": false,
|
| 11 |
+
"moe_hessian_scope": "shared",
|
| 12 |
+
"n_samples": 128,
|
| 13 |
+
"scale_zero_dtype": "float16",
|
| 14 |
+
"symmetric": false,
|
| 15 |
+
"use_error_compensation": true,
|
| 16 |
+
"use_optimal_shrinkage": true,
|
| 17 |
+
"use_weighted_quantization": true
|
| 18 |
+
},
|
| 19 |
+
"quantized_modules": {
|
| 20 |
+
"model.layers.0.mlp.down_proj": {
|
| 21 |
+
"group_size": 128,
|
| 22 |
+
"in_features": 17920,
|
| 23 |
+
"linear_dtype": "bfloat16",
|
| 24 |
+
"nbits": 4,
|
| 25 |
+
"num_groups": 716800,
|
| 26 |
+
"out_features": 5120,
|
| 27 |
+
"packing": "int4_packed_u32",
|
| 28 |
+
"quant_in_features": 17920,
|
| 29 |
+
"runtime_backend": "torch",
|
| 30 |
+
"scale_zero_dtype": "float16"
|
| 31 |
+
},
|
| 32 |
+
"model.layers.0.mlp.gate_up_proj": {
|
| 33 |
+
"group_size": 128,
|
| 34 |
+
"in_features": 5120,
|
| 35 |
+
"linear_dtype": "bfloat16",
|
| 36 |
+
"nbits": 4,
|
| 37 |
+
"num_groups": 1433600,
|
| 38 |
+
"out_features": 35840,
|
| 39 |
+
"packing": "int4_packed_u32",
|
| 40 |
+
"quant_in_features": 5120,
|
| 41 |
+
"runtime_backend": "torch",
|
| 42 |
+
"scale_zero_dtype": "float16"
|
| 43 |
+
},
|
| 44 |
+
"model.layers.0.self_attn.o_proj": {
|
| 45 |
+
"group_size": 128,
|
| 46 |
+
"in_features": 5120,
|
| 47 |
+
"linear_dtype": "bfloat16",
|
| 48 |
+
"nbits": 4,
|
| 49 |
+
"num_groups": 204800,
|
| 50 |
+
"out_features": 5120,
|
| 51 |
+
"packing": "int4_packed_u32",
|
| 52 |
+
"quant_in_features": 5120,
|
| 53 |
+
"runtime_backend": "torch",
|
| 54 |
+
"scale_zero_dtype": "float16"
|
| 55 |
+
},
|
| 56 |
+
"model.layers.0.self_attn.qkv_proj": {
|
| 57 |
+
"group_size": 128,
|
| 58 |
+
"in_features": 5120,
|
| 59 |
+
"linear_dtype": "bfloat16",
|
| 60 |
+
"nbits": 4,
|
| 61 |
+
"num_groups": 307200,
|
| 62 |
+
"out_features": 7680,
|
| 63 |
+
"packing": "int4_packed_u32",
|
| 64 |
+
"quant_in_features": 5120,
|
| 65 |
+
"runtime_backend": "torch",
|
| 66 |
+
"scale_zero_dtype": "float16"
|
| 67 |
+
},
|
| 68 |
+
"model.layers.1.mlp.down_proj": {
|
| 69 |
+
"group_size": 128,
|
| 70 |
+
"in_features": 17920,
|
| 71 |
+
"linear_dtype": "bfloat16",
|
| 72 |
+
"nbits": 4,
|
| 73 |
+
"num_groups": 716800,
|
| 74 |
+
"out_features": 5120,
|
| 75 |
+
"packing": "int4_packed_u32",
|
| 76 |
+
"quant_in_features": 17920,
|
| 77 |
+
"runtime_backend": "torch",
|
| 78 |
+
"scale_zero_dtype": "float16"
|
| 79 |
+
},
|
| 80 |
+
"model.layers.1.mlp.gate_up_proj": {
|
| 81 |
+
"group_size": 128,
|
| 82 |
+
"in_features": 5120,
|
| 83 |
+
"linear_dtype": "bfloat16",
|
| 84 |
+
"nbits": 4,
|
| 85 |
+
"num_groups": 1433600,
|
| 86 |
+
"out_features": 35840,
|
| 87 |
+
"packing": "int4_packed_u32",
|
| 88 |
+
"quant_in_features": 5120,
|
| 89 |
+
"runtime_backend": "torch",
|
| 90 |
+
"scale_zero_dtype": "float16"
|
| 91 |
+
},
|
| 92 |
+
"model.layers.1.self_attn.o_proj": {
|
| 93 |
+
"group_size": 128,
|
| 94 |
+
"in_features": 5120,
|
| 95 |
+
"linear_dtype": "bfloat16",
|
| 96 |
+
"nbits": 4,
|
| 97 |
+
"num_groups": 204800,
|
| 98 |
+
"out_features": 5120,
|
| 99 |
+
"packing": "int4_packed_u32",
|
| 100 |
+
"quant_in_features": 5120,
|
| 101 |
+
"runtime_backend": "torch",
|
| 102 |
+
"scale_zero_dtype": "float16"
|
| 103 |
+
},
|
| 104 |
+
"model.layers.1.self_attn.qkv_proj": {
|
| 105 |
+
"group_size": 128,
|
| 106 |
+
"in_features": 5120,
|
| 107 |
+
"linear_dtype": "bfloat16",
|
| 108 |
+
"nbits": 4,
|
| 109 |
+
"num_groups": 307200,
|
| 110 |
+
"out_features": 7680,
|
| 111 |
+
"packing": "int4_packed_u32",
|
| 112 |
+
"quant_in_features": 5120,
|
| 113 |
+
"runtime_backend": "torch",
|
| 114 |
+
"scale_zero_dtype": "float16"
|
| 115 |
+
},
|
| 116 |
+
"model.layers.10.mlp.down_proj": {
|
| 117 |
+
"group_size": 128,
|
| 118 |
+
"in_features": 17920,
|
| 119 |
+
"linear_dtype": "bfloat16",
|
| 120 |
+
"nbits": 4,
|
| 121 |
+
"num_groups": 716800,
|
| 122 |
+
"out_features": 5120,
|
| 123 |
+
"packing": "int4_packed_u32",
|
| 124 |
+
"quant_in_features": 17920,
|
| 125 |
+
"runtime_backend": "torch",
|
| 126 |
+
"scale_zero_dtype": "float16"
|
| 127 |
+
},
|
| 128 |
+
"model.layers.10.mlp.gate_up_proj": {
|
| 129 |
+
"group_size": 128,
|
| 130 |
+
"in_features": 5120,
|
| 131 |
+
"linear_dtype": "bfloat16",
|
| 132 |
+
"nbits": 4,
|
| 133 |
+
"num_groups": 1433600,
|
| 134 |
+
"out_features": 35840,
|
| 135 |
+
"packing": "int4_packed_u32",
|
| 136 |
+
"quant_in_features": 5120,
|
| 137 |
+
"runtime_backend": "torch",
|
| 138 |
+
"scale_zero_dtype": "float16"
|
| 139 |
+
},
|
| 140 |
+
"model.layers.10.self_attn.o_proj": {
|
| 141 |
+
"group_size": 128,
|
| 142 |
+
"in_features": 5120,
|
| 143 |
+
"linear_dtype": "bfloat16",
|
| 144 |
+
"nbits": 4,
|
| 145 |
+
"num_groups": 204800,
|
| 146 |
+
"out_features": 5120,
|
| 147 |
+
"packing": "int4_packed_u32",
|
| 148 |
+
"quant_in_features": 5120,
|
| 149 |
+
"runtime_backend": "torch",
|
| 150 |
+
"scale_zero_dtype": "float16"
|
| 151 |
+
},
|
| 152 |
+
"model.layers.10.self_attn.qkv_proj": {
|
| 153 |
+
"group_size": 128,
|
| 154 |
+
"in_features": 5120,
|
| 155 |
+
"linear_dtype": "bfloat16",
|
| 156 |
+
"nbits": 4,
|
| 157 |
+
"num_groups": 307200,
|
| 158 |
+
"out_features": 7680,
|
| 159 |
+
"packing": "int4_packed_u32",
|
| 160 |
+
"quant_in_features": 5120,
|
| 161 |
+
"runtime_backend": "torch",
|
| 162 |
+
"scale_zero_dtype": "float16"
|
| 163 |
+
},
|
| 164 |
+
"model.layers.11.mlp.down_proj": {
|
| 165 |
+
"group_size": 128,
|
| 166 |
+
"in_features": 17920,
|
| 167 |
+
"linear_dtype": "bfloat16",
|
| 168 |
+
"nbits": 4,
|
| 169 |
+
"num_groups": 716800,
|
| 170 |
+
"out_features": 5120,
|
| 171 |
+
"packing": "int4_packed_u32",
|
| 172 |
+
"quant_in_features": 17920,
|
| 173 |
+
"runtime_backend": "torch",
|
| 174 |
+
"scale_zero_dtype": "float16"
|
| 175 |
+
},
|
| 176 |
+
"model.layers.11.mlp.gate_up_proj": {
|
| 177 |
+
"group_size": 128,
|
| 178 |
+
"in_features": 5120,
|
| 179 |
+
"linear_dtype": "bfloat16",
|
| 180 |
+
"nbits": 4,
|
| 181 |
+
"num_groups": 1433600,
|
| 182 |
+
"out_features": 35840,
|
| 183 |
+
"packing": "int4_packed_u32",
|
| 184 |
+
"quant_in_features": 5120,
|
| 185 |
+
"runtime_backend": "torch",
|
| 186 |
+
"scale_zero_dtype": "float16"
|
| 187 |
+
},
|
| 188 |
+
"model.layers.11.self_attn.o_proj": {
|
| 189 |
+
"group_size": 128,
|
| 190 |
+
"in_features": 5120,
|
| 191 |
+
"linear_dtype": "bfloat16",
|
| 192 |
+
"nbits": 4,
|
| 193 |
+
"num_groups": 204800,
|
| 194 |
+
"out_features": 5120,
|
| 195 |
+
"packing": "int4_packed_u32",
|
| 196 |
+
"quant_in_features": 5120,
|
| 197 |
+
"runtime_backend": "torch",
|
| 198 |
+
"scale_zero_dtype": "float16"
|
| 199 |
+
},
|
| 200 |
+
"model.layers.11.self_attn.qkv_proj": {
|
| 201 |
+
"group_size": 128,
|
| 202 |
+
"in_features": 5120,
|
| 203 |
+
"linear_dtype": "bfloat16",
|
| 204 |
+
"nbits": 4,
|
| 205 |
+
"num_groups": 307200,
|
| 206 |
+
"out_features": 7680,
|
| 207 |
+
"packing": "int4_packed_u32",
|
| 208 |
+
"quant_in_features": 5120,
|
| 209 |
+
"runtime_backend": "torch",
|
| 210 |
+
"scale_zero_dtype": "float16"
|
| 211 |
+
},
|
| 212 |
+
"model.layers.12.mlp.down_proj": {
|
| 213 |
+
"group_size": 128,
|
| 214 |
+
"in_features": 17920,
|
| 215 |
+
"linear_dtype": "bfloat16",
|
| 216 |
+
"nbits": 4,
|
| 217 |
+
"num_groups": 716800,
|
| 218 |
+
"out_features": 5120,
|
| 219 |
+
"packing": "int4_packed_u32",
|
| 220 |
+
"quant_in_features": 17920,
|
| 221 |
+
"runtime_backend": "torch",
|
| 222 |
+
"scale_zero_dtype": "float16"
|
| 223 |
+
},
|
| 224 |
+
"model.layers.12.mlp.gate_up_proj": {
|
| 225 |
+
"group_size": 128,
|
| 226 |
+
"in_features": 5120,
|
| 227 |
+
"linear_dtype": "bfloat16",
|
| 228 |
+
"nbits": 4,
|
| 229 |
+
"num_groups": 1433600,
|
| 230 |
+
"out_features": 35840,
|
| 231 |
+
"packing": "int4_packed_u32",
|
| 232 |
+
"quant_in_features": 5120,
|
| 233 |
+
"runtime_backend": "torch",
|
| 234 |
+
"scale_zero_dtype": "float16"
|
| 235 |
+
},
|
| 236 |
+
"model.layers.12.self_attn.o_proj": {
|
| 237 |
+
"group_size": 128,
|
| 238 |
+
"in_features": 5120,
|
| 239 |
+
"linear_dtype": "bfloat16",
|
| 240 |
+
"nbits": 4,
|
| 241 |
+
"num_groups": 204800,
|
| 242 |
+
"out_features": 5120,
|
| 243 |
+
"packing": "int4_packed_u32",
|
| 244 |
+
"quant_in_features": 5120,
|
| 245 |
+
"runtime_backend": "torch",
|
| 246 |
+
"scale_zero_dtype": "float16"
|
| 247 |
+
},
|
| 248 |
+
"model.layers.12.self_attn.qkv_proj": {
|
| 249 |
+
"group_size": 128,
|
| 250 |
+
"in_features": 5120,
|
| 251 |
+
"linear_dtype": "bfloat16",
|
| 252 |
+
"nbits": 4,
|
| 253 |
+
"num_groups": 307200,
|
| 254 |
+
"out_features": 7680,
|
| 255 |
+
"packing": "int4_packed_u32",
|
| 256 |
+
"quant_in_features": 5120,
|
| 257 |
+
"runtime_backend": "torch",
|
| 258 |
+
"scale_zero_dtype": "float16"
|
| 259 |
+
},
|
| 260 |
+
"model.layers.13.mlp.down_proj": {
|
| 261 |
+
"group_size": 128,
|
| 262 |
+
"in_features": 17920,
|
| 263 |
+
"linear_dtype": "bfloat16",
|
| 264 |
+
"nbits": 4,
|
| 265 |
+
"num_groups": 716800,
|
| 266 |
+
"out_features": 5120,
|
| 267 |
+
"packing": "int4_packed_u32",
|
| 268 |
+
"quant_in_features": 17920,
|
| 269 |
+
"runtime_backend": "torch",
|
| 270 |
+
"scale_zero_dtype": "float16"
|
| 271 |
+
},
|
| 272 |
+
"model.layers.13.mlp.gate_up_proj": {
|
| 273 |
+
"group_size": 128,
|
| 274 |
+
"in_features": 5120,
|
| 275 |
+
"linear_dtype": "bfloat16",
|
| 276 |
+
"nbits": 4,
|
| 277 |
+
"num_groups": 1433600,
|
| 278 |
+
"out_features": 35840,
|
| 279 |
+
"packing": "int4_packed_u32",
|
| 280 |
+
"quant_in_features": 5120,
|
| 281 |
+
"runtime_backend": "torch",
|
| 282 |
+
"scale_zero_dtype": "float16"
|
| 283 |
+
},
|
| 284 |
+
"model.layers.13.self_attn.o_proj": {
|
| 285 |
+
"group_size": 128,
|
| 286 |
+
"in_features": 5120,
|
| 287 |
+
"linear_dtype": "bfloat16",
|
| 288 |
+
"nbits": 4,
|
| 289 |
+
"num_groups": 204800,
|
| 290 |
+
"out_features": 5120,
|
| 291 |
+
"packing": "int4_packed_u32",
|
| 292 |
+
"quant_in_features": 5120,
|
| 293 |
+
"runtime_backend": "torch",
|
| 294 |
+
"scale_zero_dtype": "float16"
|
| 295 |
+
},
|
| 296 |
+
"model.layers.13.self_attn.qkv_proj": {
|
| 297 |
+
"group_size": 128,
|
| 298 |
+
"in_features": 5120,
|
| 299 |
+
"linear_dtype": "bfloat16",
|
| 300 |
+
"nbits": 4,
|
| 301 |
+
"num_groups": 307200,
|
| 302 |
+
"out_features": 7680,
|
| 303 |
+
"packing": "int4_packed_u32",
|
| 304 |
+
"quant_in_features": 5120,
|
| 305 |
+
"runtime_backend": "torch",
|
| 306 |
+
"scale_zero_dtype": "float16"
|
| 307 |
+
},
|
| 308 |
+
"model.layers.14.mlp.down_proj": {
|
| 309 |
+
"group_size": 128,
|
| 310 |
+
"in_features": 17920,
|
| 311 |
+
"linear_dtype": "bfloat16",
|
| 312 |
+
"nbits": 4,
|
| 313 |
+
"num_groups": 716800,
|
| 314 |
+
"out_features": 5120,
|
| 315 |
+
"packing": "int4_packed_u32",
|
| 316 |
+
"quant_in_features": 17920,
|
| 317 |
+
"runtime_backend": "torch",
|
| 318 |
+
"scale_zero_dtype": "float16"
|
| 319 |
+
},
|
| 320 |
+
"model.layers.14.mlp.gate_up_proj": {
|
| 321 |
+
"group_size": 128,
|
| 322 |
+
"in_features": 5120,
|
| 323 |
+
"linear_dtype": "bfloat16",
|
| 324 |
+
"nbits": 4,
|
| 325 |
+
"num_groups": 1433600,
|
| 326 |
+
"out_features": 35840,
|
| 327 |
+
"packing": "int4_packed_u32",
|
| 328 |
+
"quant_in_features": 5120,
|
| 329 |
+
"runtime_backend": "torch",
|
| 330 |
+
"scale_zero_dtype": "float16"
|
| 331 |
+
},
|
| 332 |
+
"model.layers.14.self_attn.o_proj": {
|
| 333 |
+
"group_size": 128,
|
| 334 |
+
"in_features": 5120,
|
| 335 |
+
"linear_dtype": "bfloat16",
|
| 336 |
+
"nbits": 4,
|
| 337 |
+
"num_groups": 204800,
|
| 338 |
+
"out_features": 5120,
|
| 339 |
+
"packing": "int4_packed_u32",
|
| 340 |
+
"quant_in_features": 5120,
|
| 341 |
+
"runtime_backend": "torch",
|
| 342 |
+
"scale_zero_dtype": "float16"
|
| 343 |
+
},
|
| 344 |
+
"model.layers.14.self_attn.qkv_proj": {
|
| 345 |
+
"group_size": 128,
|
| 346 |
+
"in_features": 5120,
|
| 347 |
+
"linear_dtype": "bfloat16",
|
| 348 |
+
"nbits": 4,
|
| 349 |
+
"num_groups": 307200,
|
| 350 |
+
"out_features": 7680,
|
| 351 |
+
"packing": "int4_packed_u32",
|
| 352 |
+
"quant_in_features": 5120,
|
| 353 |
+
"runtime_backend": "torch",
|
| 354 |
+
"scale_zero_dtype": "float16"
|
| 355 |
+
},
|
| 356 |
+
"model.layers.15.mlp.down_proj": {
|
| 357 |
+
"group_size": 128,
|
| 358 |
+
"in_features": 17920,
|
| 359 |
+
"linear_dtype": "bfloat16",
|
| 360 |
+
"nbits": 4,
|
| 361 |
+
"num_groups": 716800,
|
| 362 |
+
"out_features": 5120,
|
| 363 |
+
"packing": "int4_packed_u32",
|
| 364 |
+
"quant_in_features": 17920,
|
| 365 |
+
"runtime_backend": "torch",
|
| 366 |
+
"scale_zero_dtype": "float16"
|
| 367 |
+
},
|
| 368 |
+
"model.layers.15.mlp.gate_up_proj": {
|
| 369 |
+
"group_size": 128,
|
| 370 |
+
"in_features": 5120,
|
| 371 |
+
"linear_dtype": "bfloat16",
|
| 372 |
+
"nbits": 4,
|
| 373 |
+
"num_groups": 1433600,
|
| 374 |
+
"out_features": 35840,
|
| 375 |
+
"packing": "int4_packed_u32",
|
| 376 |
+
"quant_in_features": 5120,
|
| 377 |
+
"runtime_backend": "torch",
|
| 378 |
+
"scale_zero_dtype": "float16"
|
| 379 |
+
},
|
| 380 |
+
"model.layers.15.self_attn.o_proj": {
|
| 381 |
+
"group_size": 128,
|
| 382 |
+
"in_features": 5120,
|
| 383 |
+
"linear_dtype": "bfloat16",
|
| 384 |
+
"nbits": 4,
|
| 385 |
+
"num_groups": 204800,
|
| 386 |
+
"out_features": 5120,
|
| 387 |
+
"packing": "int4_packed_u32",
|
| 388 |
+
"quant_in_features": 5120,
|
| 389 |
+
"runtime_backend": "torch",
|
| 390 |
+
"scale_zero_dtype": "float16"
|
| 391 |
+
},
|
| 392 |
+
"model.layers.15.self_attn.qkv_proj": {
|
| 393 |
+
"group_size": 128,
|
| 394 |
+
"in_features": 5120,
|
| 395 |
+
"linear_dtype": "bfloat16",
|
| 396 |
+
"nbits": 4,
|
| 397 |
+
"num_groups": 307200,
|
| 398 |
+
"out_features": 7680,
|
| 399 |
+
"packing": "int4_packed_u32",
|
| 400 |
+
"quant_in_features": 5120,
|
| 401 |
+
"runtime_backend": "torch",
|
| 402 |
+
"scale_zero_dtype": "float16"
|
| 403 |
+
},
|
| 404 |
+
"model.layers.16.mlp.down_proj": {
|
| 405 |
+
"group_size": 128,
|
| 406 |
+
"in_features": 17920,
|
| 407 |
+
"linear_dtype": "bfloat16",
|
| 408 |
+
"nbits": 4,
|
| 409 |
+
"num_groups": 716800,
|
| 410 |
+
"out_features": 5120,
|
| 411 |
+
"packing": "int4_packed_u32",
|
| 412 |
+
"quant_in_features": 17920,
|
| 413 |
+
"runtime_backend": "torch",
|
| 414 |
+
"scale_zero_dtype": "float16"
|
| 415 |
+
},
|
| 416 |
+
"model.layers.16.mlp.gate_up_proj": {
|
| 417 |
+
"group_size": 128,
|
| 418 |
+
"in_features": 5120,
|
| 419 |
+
"linear_dtype": "bfloat16",
|
| 420 |
+
"nbits": 4,
|
| 421 |
+
"num_groups": 1433600,
|
| 422 |
+
"out_features": 35840,
|
| 423 |
+
"packing": "int4_packed_u32",
|
| 424 |
+
"quant_in_features": 5120,
|
| 425 |
+
"runtime_backend": "torch",
|
| 426 |
+
"scale_zero_dtype": "float16"
|
| 427 |
+
},
|
| 428 |
+
"model.layers.16.self_attn.o_proj": {
|
| 429 |
+
"group_size": 128,
|
| 430 |
+
"in_features": 5120,
|
| 431 |
+
"linear_dtype": "bfloat16",
|
| 432 |
+
"nbits": 4,
|
| 433 |
+
"num_groups": 204800,
|
| 434 |
+
"out_features": 5120,
|
| 435 |
+
"packing": "int4_packed_u32",
|
| 436 |
+
"quant_in_features": 5120,
|
| 437 |
+
"runtime_backend": "torch",
|
| 438 |
+
"scale_zero_dtype": "float16"
|
| 439 |
+
},
|
| 440 |
+
"model.layers.16.self_attn.qkv_proj": {
|
| 441 |
+
"group_size": 128,
|
| 442 |
+
"in_features": 5120,
|
| 443 |
+
"linear_dtype": "bfloat16",
|
| 444 |
+
"nbits": 4,
|
| 445 |
+
"num_groups": 307200,
|
| 446 |
+
"out_features": 7680,
|
| 447 |
+
"packing": "int4_packed_u32",
|
| 448 |
+
"quant_in_features": 5120,
|
| 449 |
+
"runtime_backend": "torch",
|
| 450 |
+
"scale_zero_dtype": "float16"
|
| 451 |
+
},
|
| 452 |
+
"model.layers.17.mlp.down_proj": {
|
| 453 |
+
"group_size": 128,
|
| 454 |
+
"in_features": 17920,
|
| 455 |
+
"linear_dtype": "bfloat16",
|
| 456 |
+
"nbits": 4,
|
| 457 |
+
"num_groups": 716800,
|
| 458 |
+
"out_features": 5120,
|
| 459 |
+
"packing": "int4_packed_u32",
|
| 460 |
+
"quant_in_features": 17920,
|
| 461 |
+
"runtime_backend": "torch",
|
| 462 |
+
"scale_zero_dtype": "float16"
|
| 463 |
+
},
|
| 464 |
+
"model.layers.17.mlp.gate_up_proj": {
|
| 465 |
+
"group_size": 128,
|
| 466 |
+
"in_features": 5120,
|
| 467 |
+
"linear_dtype": "bfloat16",
|
| 468 |
+
"nbits": 4,
|
| 469 |
+
"num_groups": 1433600,
|
| 470 |
+
"out_features": 35840,
|
| 471 |
+
"packing": "int4_packed_u32",
|
| 472 |
+
"quant_in_features": 5120,
|
| 473 |
+
"runtime_backend": "torch",
|
| 474 |
+
"scale_zero_dtype": "float16"
|
| 475 |
+
},
|
| 476 |
+
"model.layers.17.self_attn.o_proj": {
|
| 477 |
+
"group_size": 128,
|
| 478 |
+
"in_features": 5120,
|
| 479 |
+
"linear_dtype": "bfloat16",
|
| 480 |
+
"nbits": 4,
|
| 481 |
+
"num_groups": 204800,
|
| 482 |
+
"out_features": 5120,
|
| 483 |
+
"packing": "int4_packed_u32",
|
| 484 |
+
"quant_in_features": 5120,
|
| 485 |
+
"runtime_backend": "torch",
|
| 486 |
+
"scale_zero_dtype": "float16"
|
| 487 |
+
},
|
| 488 |
+
"model.layers.17.self_attn.qkv_proj": {
|
| 489 |
+
"group_size": 128,
|
| 490 |
+
"in_features": 5120,
|
| 491 |
+
"linear_dtype": "bfloat16",
|
| 492 |
+
"nbits": 4,
|
| 493 |
+
"num_groups": 307200,
|
| 494 |
+
"out_features": 7680,
|
| 495 |
+
"packing": "int4_packed_u32",
|
| 496 |
+
"quant_in_features": 5120,
|
| 497 |
+
"runtime_backend": "torch",
|
| 498 |
+
"scale_zero_dtype": "float16"
|
| 499 |
+
},
|
| 500 |
+
"model.layers.18.mlp.down_proj": {
|
| 501 |
+
"group_size": 128,
|
| 502 |
+
"in_features": 17920,
|
| 503 |
+
"linear_dtype": "bfloat16",
|
| 504 |
+
"nbits": 4,
|
| 505 |
+
"num_groups": 716800,
|
| 506 |
+
"out_features": 5120,
|
| 507 |
+
"packing": "int4_packed_u32",
|
| 508 |
+
"quant_in_features": 17920,
|
| 509 |
+
"runtime_backend": "torch",
|
| 510 |
+
"scale_zero_dtype": "float16"
|
| 511 |
+
},
|
| 512 |
+
"model.layers.18.mlp.gate_up_proj": {
|
| 513 |
+
"group_size": 128,
|
| 514 |
+
"in_features": 5120,
|
| 515 |
+
"linear_dtype": "bfloat16",
|
| 516 |
+
"nbits": 4,
|
| 517 |
+
"num_groups": 1433600,
|
| 518 |
+
"out_features": 35840,
|
| 519 |
+
"packing": "int4_packed_u32",
|
| 520 |
+
"quant_in_features": 5120,
|
| 521 |
+
"runtime_backend": "torch",
|
| 522 |
+
"scale_zero_dtype": "float16"
|
| 523 |
+
},
|
| 524 |
+
"model.layers.18.self_attn.o_proj": {
|
| 525 |
+
"group_size": 128,
|
| 526 |
+
"in_features": 5120,
|
| 527 |
+
"linear_dtype": "bfloat16",
|
| 528 |
+
"nbits": 4,
|
| 529 |
+
"num_groups": 204800,
|
| 530 |
+
"out_features": 5120,
|
| 531 |
+
"packing": "int4_packed_u32",
|
| 532 |
+
"quant_in_features": 5120,
|
| 533 |
+
"runtime_backend": "torch",
|
| 534 |
+
"scale_zero_dtype": "float16"
|
| 535 |
+
},
|
| 536 |
+
"model.layers.18.self_attn.qkv_proj": {
|
| 537 |
+
"group_size": 128,
|
| 538 |
+
"in_features": 5120,
|
| 539 |
+
"linear_dtype": "bfloat16",
|
| 540 |
+
"nbits": 4,
|
| 541 |
+
"num_groups": 307200,
|
| 542 |
+
"out_features": 7680,
|
| 543 |
+
"packing": "int4_packed_u32",
|
| 544 |
+
"quant_in_features": 5120,
|
| 545 |
+
"runtime_backend": "torch",
|
| 546 |
+
"scale_zero_dtype": "float16"
|
| 547 |
+
},
|
| 548 |
+
"model.layers.19.mlp.down_proj": {
|
| 549 |
+
"group_size": 128,
|
| 550 |
+
"in_features": 17920,
|
| 551 |
+
"linear_dtype": "bfloat16",
|
| 552 |
+
"nbits": 4,
|
| 553 |
+
"num_groups": 716800,
|
| 554 |
+
"out_features": 5120,
|
| 555 |
+
"packing": "int4_packed_u32",
|
| 556 |
+
"quant_in_features": 17920,
|
| 557 |
+
"runtime_backend": "torch",
|
| 558 |
+
"scale_zero_dtype": "float16"
|
| 559 |
+
},
|
| 560 |
+
"model.layers.19.mlp.gate_up_proj": {
|
| 561 |
+
"group_size": 128,
|
| 562 |
+
"in_features": 5120,
|
| 563 |
+
"linear_dtype": "bfloat16",
|
| 564 |
+
"nbits": 4,
|
| 565 |
+
"num_groups": 1433600,
|
| 566 |
+
"out_features": 35840,
|
| 567 |
+
"packing": "int4_packed_u32",
|
| 568 |
+
"quant_in_features": 5120,
|
| 569 |
+
"runtime_backend": "torch",
|
| 570 |
+
"scale_zero_dtype": "float16"
|
| 571 |
+
},
|
| 572 |
+
"model.layers.19.self_attn.o_proj": {
|
| 573 |
+
"group_size": 128,
|
| 574 |
+
"in_features": 5120,
|
| 575 |
+
"linear_dtype": "bfloat16",
|
| 576 |
+
"nbits": 4,
|
| 577 |
+
"num_groups": 204800,
|
| 578 |
+
"out_features": 5120,
|
| 579 |
+
"packing": "int4_packed_u32",
|
| 580 |
+
"quant_in_features": 5120,
|
| 581 |
+
"runtime_backend": "torch",
|
| 582 |
+
"scale_zero_dtype": "float16"
|
| 583 |
+
},
|
| 584 |
+
"model.layers.19.self_attn.qkv_proj": {
|
| 585 |
+
"group_size": 128,
|
| 586 |
+
"in_features": 5120,
|
| 587 |
+
"linear_dtype": "bfloat16",
|
| 588 |
+
"nbits": 4,
|
| 589 |
+
"num_groups": 307200,
|
| 590 |
+
"out_features": 7680,
|
| 591 |
+
"packing": "int4_packed_u32",
|
| 592 |
+
"quant_in_features": 5120,
|
| 593 |
+
"runtime_backend": "torch",
|
| 594 |
+
"scale_zero_dtype": "float16"
|
| 595 |
+
},
|
| 596 |
+
"model.layers.2.mlp.down_proj": {
|
| 597 |
+
"group_size": 128,
|
| 598 |
+
"in_features": 17920,
|
| 599 |
+
"linear_dtype": "bfloat16",
|
| 600 |
+
"nbits": 4,
|
| 601 |
+
"num_groups": 716800,
|
| 602 |
+
"out_features": 5120,
|
| 603 |
+
"packing": "int4_packed_u32",
|
| 604 |
+
"quant_in_features": 17920,
|
| 605 |
+
"runtime_backend": "torch",
|
| 606 |
+
"scale_zero_dtype": "float16"
|
| 607 |
+
},
|
| 608 |
+
"model.layers.2.mlp.gate_up_proj": {
|
| 609 |
+
"group_size": 128,
|
| 610 |
+
"in_features": 5120,
|
| 611 |
+
"linear_dtype": "bfloat16",
|
| 612 |
+
"nbits": 4,
|
| 613 |
+
"num_groups": 1433600,
|
| 614 |
+
"out_features": 35840,
|
| 615 |
+
"packing": "int4_packed_u32",
|
| 616 |
+
"quant_in_features": 5120,
|
| 617 |
+
"runtime_backend": "torch",
|
| 618 |
+
"scale_zero_dtype": "float16"
|
| 619 |
+
},
|
| 620 |
+
"model.layers.2.self_attn.o_proj": {
|
| 621 |
+
"group_size": 128,
|
| 622 |
+
"in_features": 5120,
|
| 623 |
+
"linear_dtype": "bfloat16",
|
| 624 |
+
"nbits": 4,
|
| 625 |
+
"num_groups": 204800,
|
| 626 |
+
"out_features": 5120,
|
| 627 |
+
"packing": "int4_packed_u32",
|
| 628 |
+
"quant_in_features": 5120,
|
| 629 |
+
"runtime_backend": "torch",
|
| 630 |
+
"scale_zero_dtype": "float16"
|
| 631 |
+
},
|
| 632 |
+
"model.layers.2.self_attn.qkv_proj": {
|
| 633 |
+
"group_size": 128,
|
| 634 |
+
"in_features": 5120,
|
| 635 |
+
"linear_dtype": "bfloat16",
|
| 636 |
+
"nbits": 4,
|
| 637 |
+
"num_groups": 307200,
|
| 638 |
+
"out_features": 7680,
|
| 639 |
+
"packing": "int4_packed_u32",
|
| 640 |
+
"quant_in_features": 5120,
|
| 641 |
+
"runtime_backend": "torch",
|
| 642 |
+
"scale_zero_dtype": "float16"
|
| 643 |
+
},
|
| 644 |
+
"model.layers.20.mlp.down_proj": {
|
| 645 |
+
"group_size": 128,
|
| 646 |
+
"in_features": 17920,
|
| 647 |
+
"linear_dtype": "bfloat16",
|
| 648 |
+
"nbits": 4,
|
| 649 |
+
"num_groups": 716800,
|
| 650 |
+
"out_features": 5120,
|
| 651 |
+
"packing": "int4_packed_u32",
|
| 652 |
+
"quant_in_features": 17920,
|
| 653 |
+
"runtime_backend": "torch",
|
| 654 |
+
"scale_zero_dtype": "float16"
|
| 655 |
+
},
|
| 656 |
+
"model.layers.20.mlp.gate_up_proj": {
|
| 657 |
+
"group_size": 128,
|
| 658 |
+
"in_features": 5120,
|
| 659 |
+
"linear_dtype": "bfloat16",
|
| 660 |
+
"nbits": 4,
|
| 661 |
+
"num_groups": 1433600,
|
| 662 |
+
"out_features": 35840,
|
| 663 |
+
"packing": "int4_packed_u32",
|
| 664 |
+
"quant_in_features": 5120,
|
| 665 |
+
"runtime_backend": "torch",
|
| 666 |
+
"scale_zero_dtype": "float16"
|
| 667 |
+
},
|
| 668 |
+
"model.layers.20.self_attn.o_proj": {
|
| 669 |
+
"group_size": 128,
|
| 670 |
+
"in_features": 5120,
|
| 671 |
+
"linear_dtype": "bfloat16",
|
| 672 |
+
"nbits": 4,
|
| 673 |
+
"num_groups": 204800,
|
| 674 |
+
"out_features": 5120,
|
| 675 |
+
"packing": "int4_packed_u32",
|
| 676 |
+
"quant_in_features": 5120,
|
| 677 |
+
"runtime_backend": "torch",
|
| 678 |
+
"scale_zero_dtype": "float16"
|
| 679 |
+
},
|
| 680 |
+
"model.layers.20.self_attn.qkv_proj": {
|
| 681 |
+
"group_size": 128,
|
| 682 |
+
"in_features": 5120,
|
| 683 |
+
"linear_dtype": "bfloat16",
|
| 684 |
+
"nbits": 4,
|
| 685 |
+
"num_groups": 307200,
|
| 686 |
+
"out_features": 7680,
|
| 687 |
+
"packing": "int4_packed_u32",
|
| 688 |
+
"quant_in_features": 5120,
|
| 689 |
+
"runtime_backend": "torch",
|
| 690 |
+
"scale_zero_dtype": "float16"
|
| 691 |
+
},
|
| 692 |
+
"model.layers.21.mlp.down_proj": {
|
| 693 |
+
"group_size": 128,
|
| 694 |
+
"in_features": 17920,
|
| 695 |
+
"linear_dtype": "bfloat16",
|
| 696 |
+
"nbits": 4,
|
| 697 |
+
"num_groups": 716800,
|
| 698 |
+
"out_features": 5120,
|
| 699 |
+
"packing": "int4_packed_u32",
|
| 700 |
+
"quant_in_features": 17920,
|
| 701 |
+
"runtime_backend": "torch",
|
| 702 |
+
"scale_zero_dtype": "float16"
|
| 703 |
+
},
|
| 704 |
+
"model.layers.21.mlp.gate_up_proj": {
|
| 705 |
+
"group_size": 128,
|
| 706 |
+
"in_features": 5120,
|
| 707 |
+
"linear_dtype": "bfloat16",
|
| 708 |
+
"nbits": 4,
|
| 709 |
+
"num_groups": 1433600,
|
| 710 |
+
"out_features": 35840,
|
| 711 |
+
"packing": "int4_packed_u32",
|
| 712 |
+
"quant_in_features": 5120,
|
| 713 |
+
"runtime_backend": "torch",
|
| 714 |
+
"scale_zero_dtype": "float16"
|
| 715 |
+
},
|
| 716 |
+
"model.layers.21.self_attn.o_proj": {
|
| 717 |
+
"group_size": 128,
|
| 718 |
+
"in_features": 5120,
|
| 719 |
+
"linear_dtype": "bfloat16",
|
| 720 |
+
"nbits": 4,
|
| 721 |
+
"num_groups": 204800,
|
| 722 |
+
"out_features": 5120,
|
| 723 |
+
"packing": "int4_packed_u32",
|
| 724 |
+
"quant_in_features": 5120,
|
| 725 |
+
"runtime_backend": "torch",
|
| 726 |
+
"scale_zero_dtype": "float16"
|
| 727 |
+
},
|
| 728 |
+
"model.layers.21.self_attn.qkv_proj": {
|
| 729 |
+
"group_size": 128,
|
| 730 |
+
"in_features": 5120,
|
| 731 |
+
"linear_dtype": "bfloat16",
|
| 732 |
+
"nbits": 4,
|
| 733 |
+
"num_groups": 307200,
|
| 734 |
+
"out_features": 7680,
|
| 735 |
+
"packing": "int4_packed_u32",
|
| 736 |
+
"quant_in_features": 5120,
|
| 737 |
+
"runtime_backend": "torch",
|
| 738 |
+
"scale_zero_dtype": "float16"
|
| 739 |
+
},
|
| 740 |
+
"model.layers.22.mlp.down_proj": {
|
| 741 |
+
"group_size": 128,
|
| 742 |
+
"in_features": 17920,
|
| 743 |
+
"linear_dtype": "bfloat16",
|
| 744 |
+
"nbits": 4,
|
| 745 |
+
"num_groups": 716800,
|
| 746 |
+
"out_features": 5120,
|
| 747 |
+
"packing": "int4_packed_u32",
|
| 748 |
+
"quant_in_features": 17920,
|
| 749 |
+
"runtime_backend": "torch",
|
| 750 |
+
"scale_zero_dtype": "float16"
|
| 751 |
+
},
|
| 752 |
+
"model.layers.22.mlp.gate_up_proj": {
|
| 753 |
+
"group_size": 128,
|
| 754 |
+
"in_features": 5120,
|
| 755 |
+
"linear_dtype": "bfloat16",
|
| 756 |
+
"nbits": 4,
|
| 757 |
+
"num_groups": 1433600,
|
| 758 |
+
"out_features": 35840,
|
| 759 |
+
"packing": "int4_packed_u32",
|
| 760 |
+
"quant_in_features": 5120,
|
| 761 |
+
"runtime_backend": "torch",
|
| 762 |
+
"scale_zero_dtype": "float16"
|
| 763 |
+
},
|
| 764 |
+
"model.layers.22.self_attn.o_proj": {
|
| 765 |
+
"group_size": 128,
|
| 766 |
+
"in_features": 5120,
|
| 767 |
+
"linear_dtype": "bfloat16",
|
| 768 |
+
"nbits": 4,
|
| 769 |
+
"num_groups": 204800,
|
| 770 |
+
"out_features": 5120,
|
| 771 |
+
"packing": "int4_packed_u32",
|
| 772 |
+
"quant_in_features": 5120,
|
| 773 |
+
"runtime_backend": "torch",
|
| 774 |
+
"scale_zero_dtype": "float16"
|
| 775 |
+
},
|
| 776 |
+
"model.layers.22.self_attn.qkv_proj": {
|
| 777 |
+
"group_size": 128,
|
| 778 |
+
"in_features": 5120,
|
| 779 |
+
"linear_dtype": "bfloat16",
|
| 780 |
+
"nbits": 4,
|
| 781 |
+
"num_groups": 307200,
|
| 782 |
+
"out_features": 7680,
|
| 783 |
+
"packing": "int4_packed_u32",
|
| 784 |
+
"quant_in_features": 5120,
|
| 785 |
+
"runtime_backend": "torch",
|
| 786 |
+
"scale_zero_dtype": "float16"
|
| 787 |
+
},
|
| 788 |
+
"model.layers.23.mlp.down_proj": {
|
| 789 |
+
"group_size": 128,
|
| 790 |
+
"in_features": 17920,
|
| 791 |
+
"linear_dtype": "bfloat16",
|
| 792 |
+
"nbits": 4,
|
| 793 |
+
"num_groups": 716800,
|
| 794 |
+
"out_features": 5120,
|
| 795 |
+
"packing": "int4_packed_u32",
|
| 796 |
+
"quant_in_features": 17920,
|
| 797 |
+
"runtime_backend": "torch",
|
| 798 |
+
"scale_zero_dtype": "float16"
|
| 799 |
+
},
|
| 800 |
+
"model.layers.23.mlp.gate_up_proj": {
|
| 801 |
+
"group_size": 128,
|
| 802 |
+
"in_features": 5120,
|
| 803 |
+
"linear_dtype": "bfloat16",
|
| 804 |
+
"nbits": 4,
|
| 805 |
+
"num_groups": 1433600,
|
| 806 |
+
"out_features": 35840,
|
| 807 |
+
"packing": "int4_packed_u32",
|
| 808 |
+
"quant_in_features": 5120,
|
| 809 |
+
"runtime_backend": "torch",
|
| 810 |
+
"scale_zero_dtype": "float16"
|
| 811 |
+
},
|
| 812 |
+
"model.layers.23.self_attn.o_proj": {
|
| 813 |
+
"group_size": 128,
|
| 814 |
+
"in_features": 5120,
|
| 815 |
+
"linear_dtype": "bfloat16",
|
| 816 |
+
"nbits": 4,
|
| 817 |
+
"num_groups": 204800,
|
| 818 |
+
"out_features": 5120,
|
| 819 |
+
"packing": "int4_packed_u32",
|
| 820 |
+
"quant_in_features": 5120,
|
| 821 |
+
"runtime_backend": "torch",
|
| 822 |
+
"scale_zero_dtype": "float16"
|
| 823 |
+
},
|
| 824 |
+
"model.layers.23.self_attn.qkv_proj": {
|
| 825 |
+
"group_size": 128,
|
| 826 |
+
"in_features": 5120,
|
| 827 |
+
"linear_dtype": "bfloat16",
|
| 828 |
+
"nbits": 4,
|
| 829 |
+
"num_groups": 307200,
|
| 830 |
+
"out_features": 7680,
|
| 831 |
+
"packing": "int4_packed_u32",
|
| 832 |
+
"quant_in_features": 5120,
|
| 833 |
+
"runtime_backend": "torch",
|
| 834 |
+
"scale_zero_dtype": "float16"
|
| 835 |
+
},
|
| 836 |
+
"model.layers.24.mlp.down_proj": {
|
| 837 |
+
"group_size": 128,
|
| 838 |
+
"in_features": 17920,
|
| 839 |
+
"linear_dtype": "bfloat16",
|
| 840 |
+
"nbits": 4,
|
| 841 |
+
"num_groups": 716800,
|
| 842 |
+
"out_features": 5120,
|
| 843 |
+
"packing": "int4_packed_u32",
|
| 844 |
+
"quant_in_features": 17920,
|
| 845 |
+
"runtime_backend": "torch",
|
| 846 |
+
"scale_zero_dtype": "float16"
|
| 847 |
+
},
|
| 848 |
+
"model.layers.24.mlp.gate_up_proj": {
|
| 849 |
+
"group_size": 128,
|
| 850 |
+
"in_features": 5120,
|
| 851 |
+
"linear_dtype": "bfloat16",
|
| 852 |
+
"nbits": 4,
|
| 853 |
+
"num_groups": 1433600,
|
| 854 |
+
"out_features": 35840,
|
| 855 |
+
"packing": "int4_packed_u32",
|
| 856 |
+
"quant_in_features": 5120,
|
| 857 |
+
"runtime_backend": "torch",
|
| 858 |
+
"scale_zero_dtype": "float16"
|
| 859 |
+
},
|
| 860 |
+
"model.layers.24.self_attn.o_proj": {
|
| 861 |
+
"group_size": 128,
|
| 862 |
+
"in_features": 5120,
|
| 863 |
+
"linear_dtype": "bfloat16",
|
| 864 |
+
"nbits": 4,
|
| 865 |
+
"num_groups": 204800,
|
| 866 |
+
"out_features": 5120,
|
| 867 |
+
"packing": "int4_packed_u32",
|
| 868 |
+
"quant_in_features": 5120,
|
| 869 |
+
"runtime_backend": "torch",
|
| 870 |
+
"scale_zero_dtype": "float16"
|
| 871 |
+
},
|
| 872 |
+
"model.layers.24.self_attn.qkv_proj": {
|
| 873 |
+
"group_size": 128,
|
| 874 |
+
"in_features": 5120,
|
| 875 |
+
"linear_dtype": "bfloat16",
|
| 876 |
+
"nbits": 4,
|
| 877 |
+
"num_groups": 307200,
|
| 878 |
+
"out_features": 7680,
|
| 879 |
+
"packing": "int4_packed_u32",
|
| 880 |
+
"quant_in_features": 5120,
|
| 881 |
+
"runtime_backend": "torch",
|
| 882 |
+
"scale_zero_dtype": "float16"
|
| 883 |
+
},
|
| 884 |
+
"model.layers.25.mlp.down_proj": {
|
| 885 |
+
"group_size": 128,
|
| 886 |
+
"in_features": 17920,
|
| 887 |
+
"linear_dtype": "bfloat16",
|
| 888 |
+
"nbits": 4,
|
| 889 |
+
"num_groups": 716800,
|
| 890 |
+
"out_features": 5120,
|
| 891 |
+
"packing": "int4_packed_u32",
|
| 892 |
+
"quant_in_features": 17920,
|
| 893 |
+
"runtime_backend": "torch",
|
| 894 |
+
"scale_zero_dtype": "float16"
|
| 895 |
+
},
|
| 896 |
+
"model.layers.25.mlp.gate_up_proj": {
|
| 897 |
+
"group_size": 128,
|
| 898 |
+
"in_features": 5120,
|
| 899 |
+
"linear_dtype": "bfloat16",
|
| 900 |
+
"nbits": 4,
|
| 901 |
+
"num_groups": 1433600,
|
| 902 |
+
"out_features": 35840,
|
| 903 |
+
"packing": "int4_packed_u32",
|
| 904 |
+
"quant_in_features": 5120,
|
| 905 |
+
"runtime_backend": "torch",
|
| 906 |
+
"scale_zero_dtype": "float16"
|
| 907 |
+
},
|
| 908 |
+
"model.layers.25.self_attn.o_proj": {
|
| 909 |
+
"group_size": 128,
|
| 910 |
+
"in_features": 5120,
|
| 911 |
+
"linear_dtype": "bfloat16",
|
| 912 |
+
"nbits": 4,
|
| 913 |
+
"num_groups": 204800,
|
| 914 |
+
"out_features": 5120,
|
| 915 |
+
"packing": "int4_packed_u32",
|
| 916 |
+
"quant_in_features": 5120,
|
| 917 |
+
"runtime_backend": "torch",
|
| 918 |
+
"scale_zero_dtype": "float16"
|
| 919 |
+
},
|
| 920 |
+
"model.layers.25.self_attn.qkv_proj": {
|
| 921 |
+
"group_size": 128,
|
| 922 |
+
"in_features": 5120,
|
| 923 |
+
"linear_dtype": "bfloat16",
|
| 924 |
+
"nbits": 4,
|
| 925 |
+
"num_groups": 307200,
|
| 926 |
+
"out_features": 7680,
|
| 927 |
+
"packing": "int4_packed_u32",
|
| 928 |
+
"quant_in_features": 5120,
|
| 929 |
+
"runtime_backend": "torch",
|
| 930 |
+
"scale_zero_dtype": "float16"
|
| 931 |
+
},
|
| 932 |
+
"model.layers.26.mlp.down_proj": {
|
| 933 |
+
"group_size": 128,
|
| 934 |
+
"in_features": 17920,
|
| 935 |
+
"linear_dtype": "bfloat16",
|
| 936 |
+
"nbits": 4,
|
| 937 |
+
"num_groups": 716800,
|
| 938 |
+
"out_features": 5120,
|
| 939 |
+
"packing": "int4_packed_u32",
|
| 940 |
+
"quant_in_features": 17920,
|
| 941 |
+
"runtime_backend": "torch",
|
| 942 |
+
"scale_zero_dtype": "float16"
|
| 943 |
+
},
|
| 944 |
+
"model.layers.26.mlp.gate_up_proj": {
|
| 945 |
+
"group_size": 128,
|
| 946 |
+
"in_features": 5120,
|
| 947 |
+
"linear_dtype": "bfloat16",
|
| 948 |
+
"nbits": 4,
|
| 949 |
+
"num_groups": 1433600,
|
| 950 |
+
"out_features": 35840,
|
| 951 |
+
"packing": "int4_packed_u32",
|
| 952 |
+
"quant_in_features": 5120,
|
| 953 |
+
"runtime_backend": "torch",
|
| 954 |
+
"scale_zero_dtype": "float16"
|
| 955 |
+
},
|
| 956 |
+
"model.layers.26.self_attn.o_proj": {
|
| 957 |
+
"group_size": 128,
|
| 958 |
+
"in_features": 5120,
|
| 959 |
+
"linear_dtype": "bfloat16",
|
| 960 |
+
"nbits": 4,
|
| 961 |
+
"num_groups": 204800,
|
| 962 |
+
"out_features": 5120,
|
| 963 |
+
"packing": "int4_packed_u32",
|
| 964 |
+
"quant_in_features": 5120,
|
| 965 |
+
"runtime_backend": "torch",
|
| 966 |
+
"scale_zero_dtype": "float16"
|
| 967 |
+
},
|
| 968 |
+
"model.layers.26.self_attn.qkv_proj": {
|
| 969 |
+
"group_size": 128,
|
| 970 |
+
"in_features": 5120,
|
| 971 |
+
"linear_dtype": "bfloat16",
|
| 972 |
+
"nbits": 4,
|
| 973 |
+
"num_groups": 307200,
|
| 974 |
+
"out_features": 7680,
|
| 975 |
+
"packing": "int4_packed_u32",
|
| 976 |
+
"quant_in_features": 5120,
|
| 977 |
+
"runtime_backend": "torch",
|
| 978 |
+
"scale_zero_dtype": "float16"
|
| 979 |
+
},
|
| 980 |
+
"model.layers.27.mlp.down_proj": {
|
| 981 |
+
"group_size": 128,
|
| 982 |
+
"in_features": 17920,
|
| 983 |
+
"linear_dtype": "bfloat16",
|
| 984 |
+
"nbits": 4,
|
| 985 |
+
"num_groups": 716800,
|
| 986 |
+
"out_features": 5120,
|
| 987 |
+
"packing": "int4_packed_u32",
|
| 988 |
+
"quant_in_features": 17920,
|
| 989 |
+
"runtime_backend": "torch",
|
| 990 |
+
"scale_zero_dtype": "float16"
|
| 991 |
+
},
|
| 992 |
+
"model.layers.27.mlp.gate_up_proj": {
|
| 993 |
+
"group_size": 128,
|
| 994 |
+
"in_features": 5120,
|
| 995 |
+
"linear_dtype": "bfloat16",
|
| 996 |
+
"nbits": 4,
|
| 997 |
+
"num_groups": 1433600,
|
| 998 |
+
"out_features": 35840,
|
| 999 |
+
"packing": "int4_packed_u32",
|
| 1000 |
+
"quant_in_features": 5120,
|
| 1001 |
+
"runtime_backend": "torch",
|
| 1002 |
+
"scale_zero_dtype": "float16"
|
| 1003 |
+
},
|
| 1004 |
+
"model.layers.27.self_attn.o_proj": {
|
| 1005 |
+
"group_size": 128,
|
| 1006 |
+
"in_features": 5120,
|
| 1007 |
+
"linear_dtype": "bfloat16",
|
| 1008 |
+
"nbits": 4,
|
| 1009 |
+
"num_groups": 204800,
|
| 1010 |
+
"out_features": 5120,
|
| 1011 |
+
"packing": "int4_packed_u32",
|
| 1012 |
+
"quant_in_features": 5120,
|
| 1013 |
+
"runtime_backend": "torch",
|
| 1014 |
+
"scale_zero_dtype": "float16"
|
| 1015 |
+
},
|
| 1016 |
+
"model.layers.27.self_attn.qkv_proj": {
|
| 1017 |
+
"group_size": 128,
|
| 1018 |
+
"in_features": 5120,
|
| 1019 |
+
"linear_dtype": "bfloat16",
|
| 1020 |
+
"nbits": 4,
|
| 1021 |
+
"num_groups": 307200,
|
| 1022 |
+
"out_features": 7680,
|
| 1023 |
+
"packing": "int4_packed_u32",
|
| 1024 |
+
"quant_in_features": 5120,
|
| 1025 |
+
"runtime_backend": "torch",
|
| 1026 |
+
"scale_zero_dtype": "float16"
|
| 1027 |
+
},
|
| 1028 |
+
"model.layers.28.mlp.down_proj": {
|
| 1029 |
+
"group_size": 128,
|
| 1030 |
+
"in_features": 17920,
|
| 1031 |
+
"linear_dtype": "bfloat16",
|
| 1032 |
+
"nbits": 4,
|
| 1033 |
+
"num_groups": 716800,
|
| 1034 |
+
"out_features": 5120,
|
| 1035 |
+
"packing": "int4_packed_u32",
|
| 1036 |
+
"quant_in_features": 17920,
|
| 1037 |
+
"runtime_backend": "torch",
|
| 1038 |
+
"scale_zero_dtype": "float16"
|
| 1039 |
+
},
|
| 1040 |
+
"model.layers.28.mlp.gate_up_proj": {
|
| 1041 |
+
"group_size": 128,
|
| 1042 |
+
"in_features": 5120,
|
| 1043 |
+
"linear_dtype": "bfloat16",
|
| 1044 |
+
"nbits": 4,
|
| 1045 |
+
"num_groups": 1433600,
|
| 1046 |
+
"out_features": 35840,
|
| 1047 |
+
"packing": "int4_packed_u32",
|
| 1048 |
+
"quant_in_features": 5120,
|
| 1049 |
+
"runtime_backend": "torch",
|
| 1050 |
+
"scale_zero_dtype": "float16"
|
| 1051 |
+
},
|
| 1052 |
+
"model.layers.28.self_attn.o_proj": {
|
| 1053 |
+
"group_size": 128,
|
| 1054 |
+
"in_features": 5120,
|
| 1055 |
+
"linear_dtype": "bfloat16",
|
| 1056 |
+
"nbits": 4,
|
| 1057 |
+
"num_groups": 204800,
|
| 1058 |
+
"out_features": 5120,
|
| 1059 |
+
"packing": "int4_packed_u32",
|
| 1060 |
+
"quant_in_features": 5120,
|
| 1061 |
+
"runtime_backend": "torch",
|
| 1062 |
+
"scale_zero_dtype": "float16"
|
| 1063 |
+
},
|
| 1064 |
+
"model.layers.28.self_attn.qkv_proj": {
|
| 1065 |
+
"group_size": 128,
|
| 1066 |
+
"in_features": 5120,
|
| 1067 |
+
"linear_dtype": "bfloat16",
|
| 1068 |
+
"nbits": 4,
|
| 1069 |
+
"num_groups": 307200,
|
| 1070 |
+
"out_features": 7680,
|
| 1071 |
+
"packing": "int4_packed_u32",
|
| 1072 |
+
"quant_in_features": 5120,
|
| 1073 |
+
"runtime_backend": "torch",
|
| 1074 |
+
"scale_zero_dtype": "float16"
|
| 1075 |
+
},
|
| 1076 |
+
"model.layers.29.mlp.down_proj": {
|
| 1077 |
+
"group_size": 128,
|
| 1078 |
+
"in_features": 17920,
|
| 1079 |
+
"linear_dtype": "bfloat16",
|
| 1080 |
+
"nbits": 4,
|
| 1081 |
+
"num_groups": 716800,
|
| 1082 |
+
"out_features": 5120,
|
| 1083 |
+
"packing": "int4_packed_u32",
|
| 1084 |
+
"quant_in_features": 17920,
|
| 1085 |
+
"runtime_backend": "torch",
|
| 1086 |
+
"scale_zero_dtype": "float16"
|
| 1087 |
+
},
|
| 1088 |
+
"model.layers.29.mlp.gate_up_proj": {
|
| 1089 |
+
"group_size": 128,
|
| 1090 |
+
"in_features": 5120,
|
| 1091 |
+
"linear_dtype": "bfloat16",
|
| 1092 |
+
"nbits": 4,
|
| 1093 |
+
"num_groups": 1433600,
|
| 1094 |
+
"out_features": 35840,
|
| 1095 |
+
"packing": "int4_packed_u32",
|
| 1096 |
+
"quant_in_features": 5120,
|
| 1097 |
+
"runtime_backend": "torch",
|
| 1098 |
+
"scale_zero_dtype": "float16"
|
| 1099 |
+
},
|
| 1100 |
+
"model.layers.29.self_attn.o_proj": {
|
| 1101 |
+
"group_size": 128,
|
| 1102 |
+
"in_features": 5120,
|
| 1103 |
+
"linear_dtype": "bfloat16",
|
| 1104 |
+
"nbits": 4,
|
| 1105 |
+
"num_groups": 204800,
|
| 1106 |
+
"out_features": 5120,
|
| 1107 |
+
"packing": "int4_packed_u32",
|
| 1108 |
+
"quant_in_features": 5120,
|
| 1109 |
+
"runtime_backend": "torch",
|
| 1110 |
+
"scale_zero_dtype": "float16"
|
| 1111 |
+
},
|
| 1112 |
+
"model.layers.29.self_attn.qkv_proj": {
|
| 1113 |
+
"group_size": 128,
|
| 1114 |
+
"in_features": 5120,
|
| 1115 |
+
"linear_dtype": "bfloat16",
|
| 1116 |
+
"nbits": 4,
|
| 1117 |
+
"num_groups": 307200,
|
| 1118 |
+
"out_features": 7680,
|
| 1119 |
+
"packing": "int4_packed_u32",
|
| 1120 |
+
"quant_in_features": 5120,
|
| 1121 |
+
"runtime_backend": "torch",
|
| 1122 |
+
"scale_zero_dtype": "float16"
|
| 1123 |
+
},
|
| 1124 |
+
"model.layers.3.mlp.down_proj": {
|
| 1125 |
+
"group_size": 128,
|
| 1126 |
+
"in_features": 17920,
|
| 1127 |
+
"linear_dtype": "bfloat16",
|
| 1128 |
+
"nbits": 4,
|
| 1129 |
+
"num_groups": 716800,
|
| 1130 |
+
"out_features": 5120,
|
| 1131 |
+
"packing": "int4_packed_u32",
|
| 1132 |
+
"quant_in_features": 17920,
|
| 1133 |
+
"runtime_backend": "torch",
|
| 1134 |
+
"scale_zero_dtype": "float16"
|
| 1135 |
+
},
|
| 1136 |
+
"model.layers.3.mlp.gate_up_proj": {
|
| 1137 |
+
"group_size": 128,
|
| 1138 |
+
"in_features": 5120,
|
| 1139 |
+
"linear_dtype": "bfloat16",
|
| 1140 |
+
"nbits": 4,
|
| 1141 |
+
"num_groups": 1433600,
|
| 1142 |
+
"out_features": 35840,
|
| 1143 |
+
"packing": "int4_packed_u32",
|
| 1144 |
+
"quant_in_features": 5120,
|
| 1145 |
+
"runtime_backend": "torch",
|
| 1146 |
+
"scale_zero_dtype": "float16"
|
| 1147 |
+
},
|
| 1148 |
+
"model.layers.3.self_attn.o_proj": {
|
| 1149 |
+
"group_size": 128,
|
| 1150 |
+
"in_features": 5120,
|
| 1151 |
+
"linear_dtype": "bfloat16",
|
| 1152 |
+
"nbits": 4,
|
| 1153 |
+
"num_groups": 204800,
|
| 1154 |
+
"out_features": 5120,
|
| 1155 |
+
"packing": "int4_packed_u32",
|
| 1156 |
+
"quant_in_features": 5120,
|
| 1157 |
+
"runtime_backend": "torch",
|
| 1158 |
+
"scale_zero_dtype": "float16"
|
| 1159 |
+
},
|
| 1160 |
+
"model.layers.3.self_attn.qkv_proj": {
|
| 1161 |
+
"group_size": 128,
|
| 1162 |
+
"in_features": 5120,
|
| 1163 |
+
"linear_dtype": "bfloat16",
|
| 1164 |
+
"nbits": 4,
|
| 1165 |
+
"num_groups": 307200,
|
| 1166 |
+
"out_features": 7680,
|
| 1167 |
+
"packing": "int4_packed_u32",
|
| 1168 |
+
"quant_in_features": 5120,
|
| 1169 |
+
"runtime_backend": "torch",
|
| 1170 |
+
"scale_zero_dtype": "float16"
|
| 1171 |
+
},
|
| 1172 |
+
"model.layers.30.mlp.down_proj": {
|
| 1173 |
+
"group_size": 128,
|
| 1174 |
+
"in_features": 17920,
|
| 1175 |
+
"linear_dtype": "bfloat16",
|
| 1176 |
+
"nbits": 4,
|
| 1177 |
+
"num_groups": 716800,
|
| 1178 |
+
"out_features": 5120,
|
| 1179 |
+
"packing": "int4_packed_u32",
|
| 1180 |
+
"quant_in_features": 17920,
|
| 1181 |
+
"runtime_backend": "torch",
|
| 1182 |
+
"scale_zero_dtype": "float16"
|
| 1183 |
+
},
|
| 1184 |
+
"model.layers.30.mlp.gate_up_proj": {
|
| 1185 |
+
"group_size": 128,
|
| 1186 |
+
"in_features": 5120,
|
| 1187 |
+
"linear_dtype": "bfloat16",
|
| 1188 |
+
"nbits": 4,
|
| 1189 |
+
"num_groups": 1433600,
|
| 1190 |
+
"out_features": 35840,
|
| 1191 |
+
"packing": "int4_packed_u32",
|
| 1192 |
+
"quant_in_features": 5120,
|
| 1193 |
+
"runtime_backend": "torch",
|
| 1194 |
+
"scale_zero_dtype": "float16"
|
| 1195 |
+
},
|
| 1196 |
+
"model.layers.30.self_attn.o_proj": {
|
| 1197 |
+
"group_size": 128,
|
| 1198 |
+
"in_features": 5120,
|
| 1199 |
+
"linear_dtype": "bfloat16",
|
| 1200 |
+
"nbits": 4,
|
| 1201 |
+
"num_groups": 204800,
|
| 1202 |
+
"out_features": 5120,
|
| 1203 |
+
"packing": "int4_packed_u32",
|
| 1204 |
+
"quant_in_features": 5120,
|
| 1205 |
+
"runtime_backend": "torch",
|
| 1206 |
+
"scale_zero_dtype": "float16"
|
| 1207 |
+
},
|
| 1208 |
+
"model.layers.30.self_attn.qkv_proj": {
|
| 1209 |
+
"group_size": 128,
|
| 1210 |
+
"in_features": 5120,
|
| 1211 |
+
"linear_dtype": "bfloat16",
|
| 1212 |
+
"nbits": 4,
|
| 1213 |
+
"num_groups": 307200,
|
| 1214 |
+
"out_features": 7680,
|
| 1215 |
+
"packing": "int4_packed_u32",
|
| 1216 |
+
"quant_in_features": 5120,
|
| 1217 |
+
"runtime_backend": "torch",
|
| 1218 |
+
"scale_zero_dtype": "float16"
|
| 1219 |
+
},
|
| 1220 |
+
"model.layers.31.mlp.down_proj": {
|
| 1221 |
+
"group_size": 128,
|
| 1222 |
+
"in_features": 17920,
|
| 1223 |
+
"linear_dtype": "bfloat16",
|
| 1224 |
+
"nbits": 4,
|
| 1225 |
+
"num_groups": 716800,
|
| 1226 |
+
"out_features": 5120,
|
| 1227 |
+
"packing": "int4_packed_u32",
|
| 1228 |
+
"quant_in_features": 17920,
|
| 1229 |
+
"runtime_backend": "torch",
|
| 1230 |
+
"scale_zero_dtype": "float16"
|
| 1231 |
+
},
|
| 1232 |
+
"model.layers.31.mlp.gate_up_proj": {
|
| 1233 |
+
"group_size": 128,
|
| 1234 |
+
"in_features": 5120,
|
| 1235 |
+
"linear_dtype": "bfloat16",
|
| 1236 |
+
"nbits": 4,
|
| 1237 |
+
"num_groups": 1433600,
|
| 1238 |
+
"out_features": 35840,
|
| 1239 |
+
"packing": "int4_packed_u32",
|
| 1240 |
+
"quant_in_features": 5120,
|
| 1241 |
+
"runtime_backend": "torch",
|
| 1242 |
+
"scale_zero_dtype": "float16"
|
| 1243 |
+
},
|
| 1244 |
+
"model.layers.31.self_attn.o_proj": {
|
| 1245 |
+
"group_size": 128,
|
| 1246 |
+
"in_features": 5120,
|
| 1247 |
+
"linear_dtype": "bfloat16",
|
| 1248 |
+
"nbits": 4,
|
| 1249 |
+
"num_groups": 204800,
|
| 1250 |
+
"out_features": 5120,
|
| 1251 |
+
"packing": "int4_packed_u32",
|
| 1252 |
+
"quant_in_features": 5120,
|
| 1253 |
+
"runtime_backend": "torch",
|
| 1254 |
+
"scale_zero_dtype": "float16"
|
| 1255 |
+
},
|
| 1256 |
+
"model.layers.31.self_attn.qkv_proj": {
|
| 1257 |
+
"group_size": 128,
|
| 1258 |
+
"in_features": 5120,
|
| 1259 |
+
"linear_dtype": "bfloat16",
|
| 1260 |
+
"nbits": 4,
|
| 1261 |
+
"num_groups": 307200,
|
| 1262 |
+
"out_features": 7680,
|
| 1263 |
+
"packing": "int4_packed_u32",
|
| 1264 |
+
"quant_in_features": 5120,
|
| 1265 |
+
"runtime_backend": "torch",
|
| 1266 |
+
"scale_zero_dtype": "float16"
|
| 1267 |
+
},
|
| 1268 |
+
"model.layers.32.mlp.down_proj": {
|
| 1269 |
+
"group_size": 128,
|
| 1270 |
+
"in_features": 17920,
|
| 1271 |
+
"linear_dtype": "bfloat16",
|
| 1272 |
+
"nbits": 4,
|
| 1273 |
+
"num_groups": 716800,
|
| 1274 |
+
"out_features": 5120,
|
| 1275 |
+
"packing": "int4_packed_u32",
|
| 1276 |
+
"quant_in_features": 17920,
|
| 1277 |
+
"runtime_backend": "torch",
|
| 1278 |
+
"scale_zero_dtype": "float16"
|
| 1279 |
+
},
|
| 1280 |
+
"model.layers.32.mlp.gate_up_proj": {
|
| 1281 |
+
"group_size": 128,
|
| 1282 |
+
"in_features": 5120,
|
| 1283 |
+
"linear_dtype": "bfloat16",
|
| 1284 |
+
"nbits": 4,
|
| 1285 |
+
"num_groups": 1433600,
|
| 1286 |
+
"out_features": 35840,
|
| 1287 |
+
"packing": "int4_packed_u32",
|
| 1288 |
+
"quant_in_features": 5120,
|
| 1289 |
+
"runtime_backend": "torch",
|
| 1290 |
+
"scale_zero_dtype": "float16"
|
| 1291 |
+
},
|
| 1292 |
+
"model.layers.32.self_attn.o_proj": {
|
| 1293 |
+
"group_size": 128,
|
| 1294 |
+
"in_features": 5120,
|
| 1295 |
+
"linear_dtype": "bfloat16",
|
| 1296 |
+
"nbits": 4,
|
| 1297 |
+
"num_groups": 204800,
|
| 1298 |
+
"out_features": 5120,
|
| 1299 |
+
"packing": "int4_packed_u32",
|
| 1300 |
+
"quant_in_features": 5120,
|
| 1301 |
+
"runtime_backend": "torch",
|
| 1302 |
+
"scale_zero_dtype": "float16"
|
| 1303 |
+
},
|
| 1304 |
+
"model.layers.32.self_attn.qkv_proj": {
|
| 1305 |
+
"group_size": 128,
|
| 1306 |
+
"in_features": 5120,
|
| 1307 |
+
"linear_dtype": "bfloat16",
|
| 1308 |
+
"nbits": 4,
|
| 1309 |
+
"num_groups": 307200,
|
| 1310 |
+
"out_features": 7680,
|
| 1311 |
+
"packing": "int4_packed_u32",
|
| 1312 |
+
"quant_in_features": 5120,
|
| 1313 |
+
"runtime_backend": "torch",
|
| 1314 |
+
"scale_zero_dtype": "float16"
|
| 1315 |
+
},
|
| 1316 |
+
"model.layers.33.mlp.down_proj": {
|
| 1317 |
+
"group_size": 128,
|
| 1318 |
+
"in_features": 17920,
|
| 1319 |
+
"linear_dtype": "bfloat16",
|
| 1320 |
+
"nbits": 4,
|
| 1321 |
+
"num_groups": 716800,
|
| 1322 |
+
"out_features": 5120,
|
| 1323 |
+
"packing": "int4_packed_u32",
|
| 1324 |
+
"quant_in_features": 17920,
|
| 1325 |
+
"runtime_backend": "torch",
|
| 1326 |
+
"scale_zero_dtype": "float16"
|
| 1327 |
+
},
|
| 1328 |
+
"model.layers.33.mlp.gate_up_proj": {
|
| 1329 |
+
"group_size": 128,
|
| 1330 |
+
"in_features": 5120,
|
| 1331 |
+
"linear_dtype": "bfloat16",
|
| 1332 |
+
"nbits": 4,
|
| 1333 |
+
"num_groups": 1433600,
|
| 1334 |
+
"out_features": 35840,
|
| 1335 |
+
"packing": "int4_packed_u32",
|
| 1336 |
+
"quant_in_features": 5120,
|
| 1337 |
+
"runtime_backend": "torch",
|
| 1338 |
+
"scale_zero_dtype": "float16"
|
| 1339 |
+
},
|
| 1340 |
+
"model.layers.33.self_attn.o_proj": {
|
| 1341 |
+
"group_size": 128,
|
| 1342 |
+
"in_features": 5120,
|
| 1343 |
+
"linear_dtype": "bfloat16",
|
| 1344 |
+
"nbits": 4,
|
| 1345 |
+
"num_groups": 204800,
|
| 1346 |
+
"out_features": 5120,
|
| 1347 |
+
"packing": "int4_packed_u32",
|
| 1348 |
+
"quant_in_features": 5120,
|
| 1349 |
+
"runtime_backend": "torch",
|
| 1350 |
+
"scale_zero_dtype": "float16"
|
| 1351 |
+
},
|
| 1352 |
+
"model.layers.33.self_attn.qkv_proj": {
|
| 1353 |
+
"group_size": 128,
|
| 1354 |
+
"in_features": 5120,
|
| 1355 |
+
"linear_dtype": "bfloat16",
|
| 1356 |
+
"nbits": 4,
|
| 1357 |
+
"num_groups": 307200,
|
| 1358 |
+
"out_features": 7680,
|
| 1359 |
+
"packing": "int4_packed_u32",
|
| 1360 |
+
"quant_in_features": 5120,
|
| 1361 |
+
"runtime_backend": "torch",
|
| 1362 |
+
"scale_zero_dtype": "float16"
|
| 1363 |
+
},
|
| 1364 |
+
"model.layers.34.mlp.down_proj": {
|
| 1365 |
+
"group_size": 128,
|
| 1366 |
+
"in_features": 17920,
|
| 1367 |
+
"linear_dtype": "bfloat16",
|
| 1368 |
+
"nbits": 4,
|
| 1369 |
+
"num_groups": 716800,
|
| 1370 |
+
"out_features": 5120,
|
| 1371 |
+
"packing": "int4_packed_u32",
|
| 1372 |
+
"quant_in_features": 17920,
|
| 1373 |
+
"runtime_backend": "torch",
|
| 1374 |
+
"scale_zero_dtype": "float16"
|
| 1375 |
+
},
|
| 1376 |
+
"model.layers.34.mlp.gate_up_proj": {
|
| 1377 |
+
"group_size": 128,
|
| 1378 |
+
"in_features": 5120,
|
| 1379 |
+
"linear_dtype": "bfloat16",
|
| 1380 |
+
"nbits": 4,
|
| 1381 |
+
"num_groups": 1433600,
|
| 1382 |
+
"out_features": 35840,
|
| 1383 |
+
"packing": "int4_packed_u32",
|
| 1384 |
+
"quant_in_features": 5120,
|
| 1385 |
+
"runtime_backend": "torch",
|
| 1386 |
+
"scale_zero_dtype": "float16"
|
| 1387 |
+
},
|
| 1388 |
+
"model.layers.34.self_attn.o_proj": {
|
| 1389 |
+
"group_size": 128,
|
| 1390 |
+
"in_features": 5120,
|
| 1391 |
+
"linear_dtype": "bfloat16",
|
| 1392 |
+
"nbits": 4,
|
| 1393 |
+
"num_groups": 204800,
|
| 1394 |
+
"out_features": 5120,
|
| 1395 |
+
"packing": "int4_packed_u32",
|
| 1396 |
+
"quant_in_features": 5120,
|
| 1397 |
+
"runtime_backend": "torch",
|
| 1398 |
+
"scale_zero_dtype": "float16"
|
| 1399 |
+
},
|
| 1400 |
+
"model.layers.34.self_attn.qkv_proj": {
|
| 1401 |
+
"group_size": 128,
|
| 1402 |
+
"in_features": 5120,
|
| 1403 |
+
"linear_dtype": "bfloat16",
|
| 1404 |
+
"nbits": 4,
|
| 1405 |
+
"num_groups": 307200,
|
| 1406 |
+
"out_features": 7680,
|
| 1407 |
+
"packing": "int4_packed_u32",
|
| 1408 |
+
"quant_in_features": 5120,
|
| 1409 |
+
"runtime_backend": "torch",
|
| 1410 |
+
"scale_zero_dtype": "float16"
|
| 1411 |
+
},
|
| 1412 |
+
"model.layers.35.mlp.down_proj": {
|
| 1413 |
+
"group_size": 128,
|
| 1414 |
+
"in_features": 17920,
|
| 1415 |
+
"linear_dtype": "bfloat16",
|
| 1416 |
+
"nbits": 4,
|
| 1417 |
+
"num_groups": 716800,
|
| 1418 |
+
"out_features": 5120,
|
| 1419 |
+
"packing": "int4_packed_u32",
|
| 1420 |
+
"quant_in_features": 17920,
|
| 1421 |
+
"runtime_backend": "torch",
|
| 1422 |
+
"scale_zero_dtype": "float16"
|
| 1423 |
+
},
|
| 1424 |
+
"model.layers.35.mlp.gate_up_proj": {
|
| 1425 |
+
"group_size": 128,
|
| 1426 |
+
"in_features": 5120,
|
| 1427 |
+
"linear_dtype": "bfloat16",
|
| 1428 |
+
"nbits": 4,
|
| 1429 |
+
"num_groups": 1433600,
|
| 1430 |
+
"out_features": 35840,
|
| 1431 |
+
"packing": "int4_packed_u32",
|
| 1432 |
+
"quant_in_features": 5120,
|
| 1433 |
+
"runtime_backend": "torch",
|
| 1434 |
+
"scale_zero_dtype": "float16"
|
| 1435 |
+
},
|
| 1436 |
+
"model.layers.35.self_attn.o_proj": {
|
| 1437 |
+
"group_size": 128,
|
| 1438 |
+
"in_features": 5120,
|
| 1439 |
+
"linear_dtype": "bfloat16",
|
| 1440 |
+
"nbits": 4,
|
| 1441 |
+
"num_groups": 204800,
|
| 1442 |
+
"out_features": 5120,
|
| 1443 |
+
"packing": "int4_packed_u32",
|
| 1444 |
+
"quant_in_features": 5120,
|
| 1445 |
+
"runtime_backend": "torch",
|
| 1446 |
+
"scale_zero_dtype": "float16"
|
| 1447 |
+
},
|
| 1448 |
+
"model.layers.35.self_attn.qkv_proj": {
|
| 1449 |
+
"group_size": 128,
|
| 1450 |
+
"in_features": 5120,
|
| 1451 |
+
"linear_dtype": "bfloat16",
|
| 1452 |
+
"nbits": 4,
|
| 1453 |
+
"num_groups": 307200,
|
| 1454 |
+
"out_features": 7680,
|
| 1455 |
+
"packing": "int4_packed_u32",
|
| 1456 |
+
"quant_in_features": 5120,
|
| 1457 |
+
"runtime_backend": "torch",
|
| 1458 |
+
"scale_zero_dtype": "float16"
|
| 1459 |
+
},
|
| 1460 |
+
"model.layers.36.mlp.down_proj": {
|
| 1461 |
+
"group_size": 128,
|
| 1462 |
+
"in_features": 17920,
|
| 1463 |
+
"linear_dtype": "bfloat16",
|
| 1464 |
+
"nbits": 4,
|
| 1465 |
+
"num_groups": 716800,
|
| 1466 |
+
"out_features": 5120,
|
| 1467 |
+
"packing": "int4_packed_u32",
|
| 1468 |
+
"quant_in_features": 17920,
|
| 1469 |
+
"runtime_backend": "torch",
|
| 1470 |
+
"scale_zero_dtype": "float16"
|
| 1471 |
+
},
|
| 1472 |
+
"model.layers.36.mlp.gate_up_proj": {
|
| 1473 |
+
"group_size": 128,
|
| 1474 |
+
"in_features": 5120,
|
| 1475 |
+
"linear_dtype": "bfloat16",
|
| 1476 |
+
"nbits": 4,
|
| 1477 |
+
"num_groups": 1433600,
|
| 1478 |
+
"out_features": 35840,
|
| 1479 |
+
"packing": "int4_packed_u32",
|
| 1480 |
+
"quant_in_features": 5120,
|
| 1481 |
+
"runtime_backend": "torch",
|
| 1482 |
+
"scale_zero_dtype": "float16"
|
| 1483 |
+
},
|
| 1484 |
+
"model.layers.36.self_attn.o_proj": {
|
| 1485 |
+
"group_size": 128,
|
| 1486 |
+
"in_features": 5120,
|
| 1487 |
+
"linear_dtype": "bfloat16",
|
| 1488 |
+
"nbits": 4,
|
| 1489 |
+
"num_groups": 204800,
|
| 1490 |
+
"out_features": 5120,
|
| 1491 |
+
"packing": "int4_packed_u32",
|
| 1492 |
+
"quant_in_features": 5120,
|
| 1493 |
+
"runtime_backend": "torch",
|
| 1494 |
+
"scale_zero_dtype": "float16"
|
| 1495 |
+
},
|
| 1496 |
+
"model.layers.36.self_attn.qkv_proj": {
|
| 1497 |
+
"group_size": 128,
|
| 1498 |
+
"in_features": 5120,
|
| 1499 |
+
"linear_dtype": "bfloat16",
|
| 1500 |
+
"nbits": 4,
|
| 1501 |
+
"num_groups": 307200,
|
| 1502 |
+
"out_features": 7680,
|
| 1503 |
+
"packing": "int4_packed_u32",
|
| 1504 |
+
"quant_in_features": 5120,
|
| 1505 |
+
"runtime_backend": "torch",
|
| 1506 |
+
"scale_zero_dtype": "float16"
|
| 1507 |
+
},
|
| 1508 |
+
"model.layers.37.mlp.down_proj": {
|
| 1509 |
+
"group_size": 128,
|
| 1510 |
+
"in_features": 17920,
|
| 1511 |
+
"linear_dtype": "bfloat16",
|
| 1512 |
+
"nbits": 4,
|
| 1513 |
+
"num_groups": 716800,
|
| 1514 |
+
"out_features": 5120,
|
| 1515 |
+
"packing": "int4_packed_u32",
|
| 1516 |
+
"quant_in_features": 17920,
|
| 1517 |
+
"runtime_backend": "torch",
|
| 1518 |
+
"scale_zero_dtype": "float16"
|
| 1519 |
+
},
|
| 1520 |
+
"model.layers.37.mlp.gate_up_proj": {
|
| 1521 |
+
"group_size": 128,
|
| 1522 |
+
"in_features": 5120,
|
| 1523 |
+
"linear_dtype": "bfloat16",
|
| 1524 |
+
"nbits": 4,
|
| 1525 |
+
"num_groups": 1433600,
|
| 1526 |
+
"out_features": 35840,
|
| 1527 |
+
"packing": "int4_packed_u32",
|
| 1528 |
+
"quant_in_features": 5120,
|
| 1529 |
+
"runtime_backend": "torch",
|
| 1530 |
+
"scale_zero_dtype": "float16"
|
| 1531 |
+
},
|
| 1532 |
+
"model.layers.37.self_attn.o_proj": {
|
| 1533 |
+
"group_size": 128,
|
| 1534 |
+
"in_features": 5120,
|
| 1535 |
+
"linear_dtype": "bfloat16",
|
| 1536 |
+
"nbits": 4,
|
| 1537 |
+
"num_groups": 204800,
|
| 1538 |
+
"out_features": 5120,
|
| 1539 |
+
"packing": "int4_packed_u32",
|
| 1540 |
+
"quant_in_features": 5120,
|
| 1541 |
+
"runtime_backend": "torch",
|
| 1542 |
+
"scale_zero_dtype": "float16"
|
| 1543 |
+
},
|
| 1544 |
+
"model.layers.37.self_attn.qkv_proj": {
|
| 1545 |
+
"group_size": 128,
|
| 1546 |
+
"in_features": 5120,
|
| 1547 |
+
"linear_dtype": "bfloat16",
|
| 1548 |
+
"nbits": 4,
|
| 1549 |
+
"num_groups": 307200,
|
| 1550 |
+
"out_features": 7680,
|
| 1551 |
+
"packing": "int4_packed_u32",
|
| 1552 |
+
"quant_in_features": 5120,
|
| 1553 |
+
"runtime_backend": "torch",
|
| 1554 |
+
"scale_zero_dtype": "float16"
|
| 1555 |
+
},
|
| 1556 |
+
"model.layers.38.mlp.down_proj": {
|
| 1557 |
+
"group_size": 128,
|
| 1558 |
+
"in_features": 17920,
|
| 1559 |
+
"linear_dtype": "bfloat16",
|
| 1560 |
+
"nbits": 4,
|
| 1561 |
+
"num_groups": 716800,
|
| 1562 |
+
"out_features": 5120,
|
| 1563 |
+
"packing": "int4_packed_u32",
|
| 1564 |
+
"quant_in_features": 17920,
|
| 1565 |
+
"runtime_backend": "torch",
|
| 1566 |
+
"scale_zero_dtype": "float16"
|
| 1567 |
+
},
|
| 1568 |
+
"model.layers.38.mlp.gate_up_proj": {
|
| 1569 |
+
"group_size": 128,
|
| 1570 |
+
"in_features": 5120,
|
| 1571 |
+
"linear_dtype": "bfloat16",
|
| 1572 |
+
"nbits": 4,
|
| 1573 |
+
"num_groups": 1433600,
|
| 1574 |
+
"out_features": 35840,
|
| 1575 |
+
"packing": "int4_packed_u32",
|
| 1576 |
+
"quant_in_features": 5120,
|
| 1577 |
+
"runtime_backend": "torch",
|
| 1578 |
+
"scale_zero_dtype": "float16"
|
| 1579 |
+
},
|
| 1580 |
+
"model.layers.38.self_attn.o_proj": {
|
| 1581 |
+
"group_size": 128,
|
| 1582 |
+
"in_features": 5120,
|
| 1583 |
+
"linear_dtype": "bfloat16",
|
| 1584 |
+
"nbits": 4,
|
| 1585 |
+
"num_groups": 204800,
|
| 1586 |
+
"out_features": 5120,
|
| 1587 |
+
"packing": "int4_packed_u32",
|
| 1588 |
+
"quant_in_features": 5120,
|
| 1589 |
+
"runtime_backend": "torch",
|
| 1590 |
+
"scale_zero_dtype": "float16"
|
| 1591 |
+
},
|
| 1592 |
+
"model.layers.38.self_attn.qkv_proj": {
|
| 1593 |
+
"group_size": 128,
|
| 1594 |
+
"in_features": 5120,
|
| 1595 |
+
"linear_dtype": "bfloat16",
|
| 1596 |
+
"nbits": 4,
|
| 1597 |
+
"num_groups": 307200,
|
| 1598 |
+
"out_features": 7680,
|
| 1599 |
+
"packing": "int4_packed_u32",
|
| 1600 |
+
"quant_in_features": 5120,
|
| 1601 |
+
"runtime_backend": "torch",
|
| 1602 |
+
"scale_zero_dtype": "float16"
|
| 1603 |
+
},
|
| 1604 |
+
"model.layers.39.mlp.down_proj": {
|
| 1605 |
+
"group_size": 128,
|
| 1606 |
+
"in_features": 17920,
|
| 1607 |
+
"linear_dtype": "bfloat16",
|
| 1608 |
+
"nbits": 4,
|
| 1609 |
+
"num_groups": 716800,
|
| 1610 |
+
"out_features": 5120,
|
| 1611 |
+
"packing": "int4_packed_u32",
|
| 1612 |
+
"quant_in_features": 17920,
|
| 1613 |
+
"runtime_backend": "torch",
|
| 1614 |
+
"scale_zero_dtype": "float16"
|
| 1615 |
+
},
|
| 1616 |
+
"model.layers.39.mlp.gate_up_proj": {
|
| 1617 |
+
"group_size": 128,
|
| 1618 |
+
"in_features": 5120,
|
| 1619 |
+
"linear_dtype": "bfloat16",
|
| 1620 |
+
"nbits": 4,
|
| 1621 |
+
"num_groups": 1433600,
|
| 1622 |
+
"out_features": 35840,
|
| 1623 |
+
"packing": "int4_packed_u32",
|
| 1624 |
+
"quant_in_features": 5120,
|
| 1625 |
+
"runtime_backend": "torch",
|
| 1626 |
+
"scale_zero_dtype": "float16"
|
| 1627 |
+
},
|
| 1628 |
+
"model.layers.39.self_attn.o_proj": {
|
| 1629 |
+
"group_size": 128,
|
| 1630 |
+
"in_features": 5120,
|
| 1631 |
+
"linear_dtype": "bfloat16",
|
| 1632 |
+
"nbits": 4,
|
| 1633 |
+
"num_groups": 204800,
|
| 1634 |
+
"out_features": 5120,
|
| 1635 |
+
"packing": "int4_packed_u32",
|
| 1636 |
+
"quant_in_features": 5120,
|
| 1637 |
+
"runtime_backend": "torch",
|
| 1638 |
+
"scale_zero_dtype": "float16"
|
| 1639 |
+
},
|
| 1640 |
+
"model.layers.39.self_attn.qkv_proj": {
|
| 1641 |
+
"group_size": 128,
|
| 1642 |
+
"in_features": 5120,
|
| 1643 |
+
"linear_dtype": "bfloat16",
|
| 1644 |
+
"nbits": 4,
|
| 1645 |
+
"num_groups": 307200,
|
| 1646 |
+
"out_features": 7680,
|
| 1647 |
+
"packing": "int4_packed_u32",
|
| 1648 |
+
"quant_in_features": 5120,
|
| 1649 |
+
"runtime_backend": "torch",
|
| 1650 |
+
"scale_zero_dtype": "float16"
|
| 1651 |
+
},
|
| 1652 |
+
"model.layers.4.mlp.down_proj": {
|
| 1653 |
+
"group_size": 128,
|
| 1654 |
+
"in_features": 17920,
|
| 1655 |
+
"linear_dtype": "bfloat16",
|
| 1656 |
+
"nbits": 4,
|
| 1657 |
+
"num_groups": 716800,
|
| 1658 |
+
"out_features": 5120,
|
| 1659 |
+
"packing": "int4_packed_u32",
|
| 1660 |
+
"quant_in_features": 17920,
|
| 1661 |
+
"runtime_backend": "torch",
|
| 1662 |
+
"scale_zero_dtype": "float16"
|
| 1663 |
+
},
|
| 1664 |
+
"model.layers.4.mlp.gate_up_proj": {
|
| 1665 |
+
"group_size": 128,
|
| 1666 |
+
"in_features": 5120,
|
| 1667 |
+
"linear_dtype": "bfloat16",
|
| 1668 |
+
"nbits": 4,
|
| 1669 |
+
"num_groups": 1433600,
|
| 1670 |
+
"out_features": 35840,
|
| 1671 |
+
"packing": "int4_packed_u32",
|
| 1672 |
+
"quant_in_features": 5120,
|
| 1673 |
+
"runtime_backend": "torch",
|
| 1674 |
+
"scale_zero_dtype": "float16"
|
| 1675 |
+
},
|
| 1676 |
+
"model.layers.4.self_attn.o_proj": {
|
| 1677 |
+
"group_size": 128,
|
| 1678 |
+
"in_features": 5120,
|
| 1679 |
+
"linear_dtype": "bfloat16",
|
| 1680 |
+
"nbits": 4,
|
| 1681 |
+
"num_groups": 204800,
|
| 1682 |
+
"out_features": 5120,
|
| 1683 |
+
"packing": "int4_packed_u32",
|
| 1684 |
+
"quant_in_features": 5120,
|
| 1685 |
+
"runtime_backend": "torch",
|
| 1686 |
+
"scale_zero_dtype": "float16"
|
| 1687 |
+
},
|
| 1688 |
+
"model.layers.4.self_attn.qkv_proj": {
|
| 1689 |
+
"group_size": 128,
|
| 1690 |
+
"in_features": 5120,
|
| 1691 |
+
"linear_dtype": "bfloat16",
|
| 1692 |
+
"nbits": 4,
|
| 1693 |
+
"num_groups": 307200,
|
| 1694 |
+
"out_features": 7680,
|
| 1695 |
+
"packing": "int4_packed_u32",
|
| 1696 |
+
"quant_in_features": 5120,
|
| 1697 |
+
"runtime_backend": "torch",
|
| 1698 |
+
"scale_zero_dtype": "float16"
|
| 1699 |
+
},
|
| 1700 |
+
"model.layers.5.mlp.down_proj": {
|
| 1701 |
+
"group_size": 128,
|
| 1702 |
+
"in_features": 17920,
|
| 1703 |
+
"linear_dtype": "bfloat16",
|
| 1704 |
+
"nbits": 4,
|
| 1705 |
+
"num_groups": 716800,
|
| 1706 |
+
"out_features": 5120,
|
| 1707 |
+
"packing": "int4_packed_u32",
|
| 1708 |
+
"quant_in_features": 17920,
|
| 1709 |
+
"runtime_backend": "torch",
|
| 1710 |
+
"scale_zero_dtype": "float16"
|
| 1711 |
+
},
|
| 1712 |
+
"model.layers.5.mlp.gate_up_proj": {
|
| 1713 |
+
"group_size": 128,
|
| 1714 |
+
"in_features": 5120,
|
| 1715 |
+
"linear_dtype": "bfloat16",
|
| 1716 |
+
"nbits": 4,
|
| 1717 |
+
"num_groups": 1433600,
|
| 1718 |
+
"out_features": 35840,
|
| 1719 |
+
"packing": "int4_packed_u32",
|
| 1720 |
+
"quant_in_features": 5120,
|
| 1721 |
+
"runtime_backend": "torch",
|
| 1722 |
+
"scale_zero_dtype": "float16"
|
| 1723 |
+
},
|
| 1724 |
+
"model.layers.5.self_attn.o_proj": {
|
| 1725 |
+
"group_size": 128,
|
| 1726 |
+
"in_features": 5120,
|
| 1727 |
+
"linear_dtype": "bfloat16",
|
| 1728 |
+
"nbits": 4,
|
| 1729 |
+
"num_groups": 204800,
|
| 1730 |
+
"out_features": 5120,
|
| 1731 |
+
"packing": "int4_packed_u32",
|
| 1732 |
+
"quant_in_features": 5120,
|
| 1733 |
+
"runtime_backend": "torch",
|
| 1734 |
+
"scale_zero_dtype": "float16"
|
| 1735 |
+
},
|
| 1736 |
+
"model.layers.5.self_attn.qkv_proj": {
|
| 1737 |
+
"group_size": 128,
|
| 1738 |
+
"in_features": 5120,
|
| 1739 |
+
"linear_dtype": "bfloat16",
|
| 1740 |
+
"nbits": 4,
|
| 1741 |
+
"num_groups": 307200,
|
| 1742 |
+
"out_features": 7680,
|
| 1743 |
+
"packing": "int4_packed_u32",
|
| 1744 |
+
"quant_in_features": 5120,
|
| 1745 |
+
"runtime_backend": "torch",
|
| 1746 |
+
"scale_zero_dtype": "float16"
|
| 1747 |
+
},
|
| 1748 |
+
"model.layers.6.mlp.down_proj": {
|
| 1749 |
+
"group_size": 128,
|
| 1750 |
+
"in_features": 17920,
|
| 1751 |
+
"linear_dtype": "bfloat16",
|
| 1752 |
+
"nbits": 4,
|
| 1753 |
+
"num_groups": 716800,
|
| 1754 |
+
"out_features": 5120,
|
| 1755 |
+
"packing": "int4_packed_u32",
|
| 1756 |
+
"quant_in_features": 17920,
|
| 1757 |
+
"runtime_backend": "torch",
|
| 1758 |
+
"scale_zero_dtype": "float16"
|
| 1759 |
+
},
|
| 1760 |
+
"model.layers.6.mlp.gate_up_proj": {
|
| 1761 |
+
"group_size": 128,
|
| 1762 |
+
"in_features": 5120,
|
| 1763 |
+
"linear_dtype": "bfloat16",
|
| 1764 |
+
"nbits": 4,
|
| 1765 |
+
"num_groups": 1433600,
|
| 1766 |
+
"out_features": 35840,
|
| 1767 |
+
"packing": "int4_packed_u32",
|
| 1768 |
+
"quant_in_features": 5120,
|
| 1769 |
+
"runtime_backend": "torch",
|
| 1770 |
+
"scale_zero_dtype": "float16"
|
| 1771 |
+
},
|
| 1772 |
+
"model.layers.6.self_attn.o_proj": {
|
| 1773 |
+
"group_size": 128,
|
| 1774 |
+
"in_features": 5120,
|
| 1775 |
+
"linear_dtype": "bfloat16",
|
| 1776 |
+
"nbits": 4,
|
| 1777 |
+
"num_groups": 204800,
|
| 1778 |
+
"out_features": 5120,
|
| 1779 |
+
"packing": "int4_packed_u32",
|
| 1780 |
+
"quant_in_features": 5120,
|
| 1781 |
+
"runtime_backend": "torch",
|
| 1782 |
+
"scale_zero_dtype": "float16"
|
| 1783 |
+
},
|
| 1784 |
+
"model.layers.6.self_attn.qkv_proj": {
|
| 1785 |
+
"group_size": 128,
|
| 1786 |
+
"in_features": 5120,
|
| 1787 |
+
"linear_dtype": "bfloat16",
|
| 1788 |
+
"nbits": 4,
|
| 1789 |
+
"num_groups": 307200,
|
| 1790 |
+
"out_features": 7680,
|
| 1791 |
+
"packing": "int4_packed_u32",
|
| 1792 |
+
"quant_in_features": 5120,
|
| 1793 |
+
"runtime_backend": "torch",
|
| 1794 |
+
"scale_zero_dtype": "float16"
|
| 1795 |
+
},
|
| 1796 |
+
"model.layers.7.mlp.down_proj": {
|
| 1797 |
+
"group_size": 128,
|
| 1798 |
+
"in_features": 17920,
|
| 1799 |
+
"linear_dtype": "bfloat16",
|
| 1800 |
+
"nbits": 4,
|
| 1801 |
+
"num_groups": 716800,
|
| 1802 |
+
"out_features": 5120,
|
| 1803 |
+
"packing": "int4_packed_u32",
|
| 1804 |
+
"quant_in_features": 17920,
|
| 1805 |
+
"runtime_backend": "torch",
|
| 1806 |
+
"scale_zero_dtype": "float16"
|
| 1807 |
+
},
|
| 1808 |
+
"model.layers.7.mlp.gate_up_proj": {
|
| 1809 |
+
"group_size": 128,
|
| 1810 |
+
"in_features": 5120,
|
| 1811 |
+
"linear_dtype": "bfloat16",
|
| 1812 |
+
"nbits": 4,
|
| 1813 |
+
"num_groups": 1433600,
|
| 1814 |
+
"out_features": 35840,
|
| 1815 |
+
"packing": "int4_packed_u32",
|
| 1816 |
+
"quant_in_features": 5120,
|
| 1817 |
+
"runtime_backend": "torch",
|
| 1818 |
+
"scale_zero_dtype": "float16"
|
| 1819 |
+
},
|
| 1820 |
+
"model.layers.7.self_attn.o_proj": {
|
| 1821 |
+
"group_size": 128,
|
| 1822 |
+
"in_features": 5120,
|
| 1823 |
+
"linear_dtype": "bfloat16",
|
| 1824 |
+
"nbits": 4,
|
| 1825 |
+
"num_groups": 204800,
|
| 1826 |
+
"out_features": 5120,
|
| 1827 |
+
"packing": "int4_packed_u32",
|
| 1828 |
+
"quant_in_features": 5120,
|
| 1829 |
+
"runtime_backend": "torch",
|
| 1830 |
+
"scale_zero_dtype": "float16"
|
| 1831 |
+
},
|
| 1832 |
+
"model.layers.7.self_attn.qkv_proj": {
|
| 1833 |
+
"group_size": 128,
|
| 1834 |
+
"in_features": 5120,
|
| 1835 |
+
"linear_dtype": "bfloat16",
|
| 1836 |
+
"nbits": 4,
|
| 1837 |
+
"num_groups": 307200,
|
| 1838 |
+
"out_features": 7680,
|
| 1839 |
+
"packing": "int4_packed_u32",
|
| 1840 |
+
"quant_in_features": 5120,
|
| 1841 |
+
"runtime_backend": "torch",
|
| 1842 |
+
"scale_zero_dtype": "float16"
|
| 1843 |
+
},
|
| 1844 |
+
"model.layers.8.mlp.down_proj": {
|
| 1845 |
+
"group_size": 128,
|
| 1846 |
+
"in_features": 17920,
|
| 1847 |
+
"linear_dtype": "bfloat16",
|
| 1848 |
+
"nbits": 4,
|
| 1849 |
+
"num_groups": 716800,
|
| 1850 |
+
"out_features": 5120,
|
| 1851 |
+
"packing": "int4_packed_u32",
|
| 1852 |
+
"quant_in_features": 17920,
|
| 1853 |
+
"runtime_backend": "torch",
|
| 1854 |
+
"scale_zero_dtype": "float16"
|
| 1855 |
+
},
|
| 1856 |
+
"model.layers.8.mlp.gate_up_proj": {
|
| 1857 |
+
"group_size": 128,
|
| 1858 |
+
"in_features": 5120,
|
| 1859 |
+
"linear_dtype": "bfloat16",
|
| 1860 |
+
"nbits": 4,
|
| 1861 |
+
"num_groups": 1433600,
|
| 1862 |
+
"out_features": 35840,
|
| 1863 |
+
"packing": "int4_packed_u32",
|
| 1864 |
+
"quant_in_features": 5120,
|
| 1865 |
+
"runtime_backend": "torch",
|
| 1866 |
+
"scale_zero_dtype": "float16"
|
| 1867 |
+
},
|
| 1868 |
+
"model.layers.8.self_attn.o_proj": {
|
| 1869 |
+
"group_size": 128,
|
| 1870 |
+
"in_features": 5120,
|
| 1871 |
+
"linear_dtype": "bfloat16",
|
| 1872 |
+
"nbits": 4,
|
| 1873 |
+
"num_groups": 204800,
|
| 1874 |
+
"out_features": 5120,
|
| 1875 |
+
"packing": "int4_packed_u32",
|
| 1876 |
+
"quant_in_features": 5120,
|
| 1877 |
+
"runtime_backend": "torch",
|
| 1878 |
+
"scale_zero_dtype": "float16"
|
| 1879 |
+
},
|
| 1880 |
+
"model.layers.8.self_attn.qkv_proj": {
|
| 1881 |
+
"group_size": 128,
|
| 1882 |
+
"in_features": 5120,
|
| 1883 |
+
"linear_dtype": "bfloat16",
|
| 1884 |
+
"nbits": 4,
|
| 1885 |
+
"num_groups": 307200,
|
| 1886 |
+
"out_features": 7680,
|
| 1887 |
+
"packing": "int4_packed_u32",
|
| 1888 |
+
"quant_in_features": 5120,
|
| 1889 |
+
"runtime_backend": "torch",
|
| 1890 |
+
"scale_zero_dtype": "float16"
|
| 1891 |
+
},
|
| 1892 |
+
"model.layers.9.mlp.down_proj": {
|
| 1893 |
+
"group_size": 128,
|
| 1894 |
+
"in_features": 17920,
|
| 1895 |
+
"linear_dtype": "bfloat16",
|
| 1896 |
+
"nbits": 4,
|
| 1897 |
+
"num_groups": 716800,
|
| 1898 |
+
"out_features": 5120,
|
| 1899 |
+
"packing": "int4_packed_u32",
|
| 1900 |
+
"quant_in_features": 17920,
|
| 1901 |
+
"runtime_backend": "torch",
|
| 1902 |
+
"scale_zero_dtype": "float16"
|
| 1903 |
+
},
|
| 1904 |
+
"model.layers.9.mlp.gate_up_proj": {
|
| 1905 |
+
"group_size": 128,
|
| 1906 |
+
"in_features": 5120,
|
| 1907 |
+
"linear_dtype": "bfloat16",
|
| 1908 |
+
"nbits": 4,
|
| 1909 |
+
"num_groups": 1433600,
|
| 1910 |
+
"out_features": 35840,
|
| 1911 |
+
"packing": "int4_packed_u32",
|
| 1912 |
+
"quant_in_features": 5120,
|
| 1913 |
+
"runtime_backend": "torch",
|
| 1914 |
+
"scale_zero_dtype": "float16"
|
| 1915 |
+
},
|
| 1916 |
+
"model.layers.9.self_attn.o_proj": {
|
| 1917 |
+
"group_size": 128,
|
| 1918 |
+
"in_features": 5120,
|
| 1919 |
+
"linear_dtype": "bfloat16",
|
| 1920 |
+
"nbits": 4,
|
| 1921 |
+
"num_groups": 204800,
|
| 1922 |
+
"out_features": 5120,
|
| 1923 |
+
"packing": "int4_packed_u32",
|
| 1924 |
+
"quant_in_features": 5120,
|
| 1925 |
+
"runtime_backend": "torch",
|
| 1926 |
+
"scale_zero_dtype": "float16"
|
| 1927 |
+
},
|
| 1928 |
+
"model.layers.9.self_attn.qkv_proj": {
|
| 1929 |
+
"group_size": 128,
|
| 1930 |
+
"in_features": 5120,
|
| 1931 |
+
"linear_dtype": "bfloat16",
|
| 1932 |
+
"nbits": 4,
|
| 1933 |
+
"num_groups": 307200,
|
| 1934 |
+
"out_features": 7680,
|
| 1935 |
+
"packing": "int4_packed_u32",
|
| 1936 |
+
"quant_in_features": 5120,
|
| 1937 |
+
"runtime_backend": "torch",
|
| 1938 |
+
"scale_zero_dtype": "float16"
|
| 1939 |
+
}
|
| 1940 |
+
},
|
| 1941 |
+
"results": {
|
| 1942 |
+
"Method": "dashq",
|
| 1943 |
+
"Model": "microsoft/phi-4",
|
| 1944 |
+
"ModelSizeGB": 9.297831568,
|
| 1945 |
+
"OriginalSizeGB": 29.319042992,
|
| 1946 |
+
"PPL": 6.515074253082275,
|
| 1947 |
+
"Params": "{'bits': 4, 'group_size': 128, 'scale_zero_dtype': 'float16', 'n_samples': 128, 'moe_hessian_scope': 'shared', 'use_error_compensation': True, 'use_optimal_shrinkage': True, 'use_weighted_quantization': True, 'symmetric': False, 'low_memory_optimization': False}",
|
| 1948 |
+
"QuantTime": 994.4847304821014,
|
| 1949 |
+
"arc_challenge": 56.22866894197952,
|
| 1950 |
+
"arc_easy": 74.32659932659934,
|
| 1951 |
+
"commonsense_qa": 74.44717444717445,
|
| 1952 |
+
"hellaswag": 81.84624576777534,
|
| 1953 |
+
"lambada_openai": 72.28798758005045,
|
| 1954 |
+
"openbookqa": 45.4,
|
| 1955 |
+
"piqa": 81.33841131664853,
|
| 1956 |
+
"truthfulqa_mc2": 59.33316399827141,
|
| 1957 |
+
"winogrande": 76.16416732438832,
|
| 1958 |
+
"zeroshot_avg": 69.04137985587637
|
| 1959 |
+
}
|
| 1960 |
+
}
|
dashq_kernel.py
ADDED
|
@@ -0,0 +1,301 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Triton kernels for group-wise asymmetric integer weights.
|
| 2 |
+
|
| 3 |
+
Weights are stored K-major: W_q has shape (K // elements_per_word, N) with
|
| 4 |
+
values packed along K, and scale/zero have shape (K // group_size, N).
|
| 5 |
+
|
| 6 |
+
Three kernels are selected by the number of input rows M:
|
| 7 |
+
|
| 8 |
+
M == 1 GEMV
|
| 9 |
+
2 <= M <= 32 fused dequantize-GEMM with split-K
|
| 10 |
+
M > 32 fused dequantize-GEMM, accumulator kept in registers
|
| 11 |
+
|
| 12 |
+
Supported bit widths are 1, 2, 3, 4 and 8. 3-bit is stored as a 2-bit plane
|
| 13 |
+
plus a 1-bit plane, so it occupies exactly 3 bits per weight.
|
| 14 |
+
"""
|
| 15 |
+
from __future__ import annotations
|
| 16 |
+
|
| 17 |
+
from typing import Optional
|
| 18 |
+
|
| 19 |
+
import torch
|
| 20 |
+
import torch.nn as nn
|
| 21 |
+
|
| 22 |
+
try:
|
| 23 |
+
import triton
|
| 24 |
+
import triton.language as tl
|
| 25 |
+
|
| 26 |
+
TRITON_AVAILABLE = True
|
| 27 |
+
except Exception: # triton is optional
|
| 28 |
+
TRITON_AVAILABLE = False
|
| 29 |
+
|
| 30 |
+
SUPPORTED_NBITS = (1, 2, 3, 4, 8)
|
| 31 |
+
|
| 32 |
+
_GEMM_CONFIG_CACHE = {}
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
if TRITON_AVAILABLE:
|
| 36 |
+
|
| 37 |
+
@triton.jit
|
| 38 |
+
def _dashq_gemv_kernel(
|
| 39 |
+
x_ptr, w_ptr, lo_ptr, s_ptr, z_ptr, y_ptr,
|
| 40 |
+
N, K,
|
| 41 |
+
NBITS: tl.constexpr, EPS: tl.constexpr, GS: tl.constexpr,
|
| 42 |
+
BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr,
|
| 43 |
+
):
|
| 44 |
+
pid_n = tl.program_id(0)
|
| 45 |
+
pid_k = tl.program_id(1) * 2
|
| 46 |
+
offs_n = pid_n * BLOCK_N + tl.arange(0, BLOCK_N)
|
| 47 |
+
offs_n = tl.max_contiguous(tl.multiple_of(offs_n, BLOCK_N), BLOCK_N)
|
| 48 |
+
|
| 49 |
+
# 2 * BLOCK_K == GS, so a program covers exactly one scale group.
|
| 50 |
+
k_m = (pid_k * BLOCK_K) // GS
|
| 51 |
+
scales = tl.load(s_ptr + k_m * N + offs_n).to(tl.float32)
|
| 52 |
+
zeros = tl.load(z_ptr + k_m * N + offs_n).to(tl.float32)
|
| 53 |
+
|
| 54 |
+
acc = tl.zeros((BLOCK_N,), dtype=tl.float32)
|
| 55 |
+
offs_k = pid_k * BLOCK_K + tl.arange(0, BLOCK_K)
|
| 56 |
+
for _ in tl.static_range(2):
|
| 57 |
+
a = tl.load(x_ptr + offs_k, eviction_policy="evict_last").to(tl.float32)
|
| 58 |
+
if NBITS == 3:
|
| 59 |
+
hw = tl.load(
|
| 60 |
+
w_ptr + (offs_k // 16)[:, None] * N + offs_n[None, :],
|
| 61 |
+
eviction_policy="evict_first",
|
| 62 |
+
)
|
| 63 |
+
lw = tl.load(
|
| 64 |
+
lo_ptr + (offs_k // 32)[:, None] * N + offs_n[None, :],
|
| 65 |
+
eviction_policy="evict_first",
|
| 66 |
+
)
|
| 67 |
+
q = (((hw >> (((offs_k % 16) * 2)[:, None])) & 3) << 1) | (
|
| 68 |
+
(lw >> ((offs_k % 32)[:, None])) & 1
|
| 69 |
+
)
|
| 70 |
+
else:
|
| 71 |
+
wv = tl.load(
|
| 72 |
+
w_ptr + (offs_k // EPS)[:, None] * N + offs_n[None, :],
|
| 73 |
+
eviction_policy="evict_first",
|
| 74 |
+
)
|
| 75 |
+
q = (wv >> (((offs_k % EPS) * NBITS)[:, None])) & ((1 << NBITS) - 1)
|
| 76 |
+
b = (q.to(tl.float32) - zeros[None, :]) * scales[None, :]
|
| 77 |
+
acc += tl.sum(a[:, None] * b, axis=0)
|
| 78 |
+
offs_k += BLOCK_K
|
| 79 |
+
tl.atomic_add(y_ptr + offs_n, acc, sem="relaxed")
|
| 80 |
+
|
| 81 |
+
@triton.jit
|
| 82 |
+
def _dashq_gemm_kernel(
|
| 83 |
+
x_ptr, w_ptr, lo_ptr, s_ptr, z_ptr, y_ptr,
|
| 84 |
+
M, N, K,
|
| 85 |
+
NBITS: tl.constexpr, EPS: tl.constexpr, GS: tl.constexpr,
|
| 86 |
+
BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr,
|
| 87 |
+
SPLIT_K: tl.constexpr, OUT_DTYPE: tl.constexpr,
|
| 88 |
+
):
|
| 89 |
+
"""y[M, N] = x[M, K] @ dequantize(w)[K, N]
|
| 90 |
+
|
| 91 |
+
BLOCK_K divides the group size, so a K-tile lies inside one group and the
|
| 92 |
+
scale/zero load is a single (1, BLOCK_N) vector.
|
| 93 |
+
"""
|
| 94 |
+
pid_m = tl.program_id(0)
|
| 95 |
+
pid_n = tl.program_id(1)
|
| 96 |
+
pid_k = tl.program_id(2)
|
| 97 |
+
|
| 98 |
+
offs_m = pid_m * BLOCK_M + tl.arange(0, BLOCK_M)
|
| 99 |
+
offs_n = pid_n * BLOCK_N + tl.arange(0, BLOCK_N)
|
| 100 |
+
mask_m = offs_m < M
|
| 101 |
+
mask_n = offs_n < N
|
| 102 |
+
|
| 103 |
+
acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=tl.float32)
|
| 104 |
+
|
| 105 |
+
for t in range(pid_k, tl.cdiv(K, BLOCK_K), SPLIT_K):
|
| 106 |
+
k0 = t * BLOCK_K
|
| 107 |
+
offs_k = k0 + tl.arange(0, BLOCK_K)
|
| 108 |
+
mask_k = offs_k < K
|
| 109 |
+
|
| 110 |
+
x = tl.load(x_ptr + offs_m[:, None] * K + offs_k[None, :],
|
| 111 |
+
mask=mask_m[:, None] & mask_k[None, :], other=0.0)
|
| 112 |
+
|
| 113 |
+
if NBITS == 3:
|
| 114 |
+
hw = tl.load(w_ptr + (offs_k // 16)[:, None] * N + offs_n[None, :],
|
| 115 |
+
mask=mask_k[:, None] & mask_n[None, :], other=0)
|
| 116 |
+
lw = tl.load(lo_ptr + (offs_k // 32)[:, None] * N + offs_n[None, :],
|
| 117 |
+
mask=mask_k[:, None] & mask_n[None, :], other=0)
|
| 118 |
+
q = (((hw >> (((offs_k % 16) * 2)[:, None])) & 3) << 1) | (
|
| 119 |
+
(lw >> ((offs_k % 32)[:, None])) & 1)
|
| 120 |
+
else:
|
| 121 |
+
wv = tl.load(w_ptr + (offs_k // EPS)[:, None] * N + offs_n[None, :],
|
| 122 |
+
mask=mask_k[:, None] & mask_n[None, :], other=0)
|
| 123 |
+
q = (wv >> (((offs_k % EPS) * NBITS)[:, None])) & ((1 << NBITS) - 1)
|
| 124 |
+
|
| 125 |
+
g = k0 // GS
|
| 126 |
+
s = tl.load(s_ptr + g * N + offs_n, mask=mask_n, other=0.0).to(tl.float32)
|
| 127 |
+
z = tl.load(z_ptr + g * N + offs_n, mask=mask_n, other=0.0).to(tl.float32)
|
| 128 |
+
w = (q.to(tl.float32) - z[None, :]) * s[None, :]
|
| 129 |
+
|
| 130 |
+
acc += tl.dot(x, w.to(x.dtype), out_dtype=tl.float32)
|
| 131 |
+
|
| 132 |
+
out = acc.to(OUT_DTYPE)
|
| 133 |
+
y_ptrs = y_ptr + offs_m[:, None] * N + offs_n[None, :]
|
| 134 |
+
if SPLIT_K == 1:
|
| 135 |
+
tl.store(y_ptrs, out, mask=mask_m[:, None] & mask_n[None, :])
|
| 136 |
+
else:
|
| 137 |
+
tl.atomic_add(y_ptrs, out, mask=mask_m[:, None] & mask_n[None, :], sem="relaxed")
|
| 138 |
+
|
| 139 |
+
|
| 140 |
+
def _pack_kmajor(q_kn: torch.Tensor, bits: int) -> torch.Tensor:
|
| 141 |
+
"""(K, N) codes -> (K // eps, N) int32, value k in word k // eps."""
|
| 142 |
+
K, N = q_kn.shape
|
| 143 |
+
eps = 32 // bits
|
| 144 |
+
v = q_kn.to(torch.int32).reshape(K // eps, eps, N)
|
| 145 |
+
words = torch.zeros(K // eps, N, dtype=torch.int32, device=q_kn.device)
|
| 146 |
+
for j in range(eps):
|
| 147 |
+
words |= v[:, j, :] << (bits * j)
|
| 148 |
+
return words
|
| 149 |
+
|
| 150 |
+
|
| 151 |
+
def _unpack_kmajor(words: torch.Tensor, bits: int, K: int) -> torch.Tensor:
|
| 152 |
+
eps = 32 // bits
|
| 153 |
+
WK, N = words.shape
|
| 154 |
+
shifts = (torch.arange(eps, device=words.device, dtype=torch.int32) * bits).view(1, eps, 1)
|
| 155 |
+
q = (words.view(WK, 1, N) >> shifts) & ((1 << bits) - 1)
|
| 156 |
+
return q.reshape(WK * eps, N)[:K]
|
| 157 |
+
|
| 158 |
+
|
| 159 |
+
class TritonQuantLinear(nn.Module):
|
| 160 |
+
"""Linear layer over group-wise asymmetric integer weights."""
|
| 161 |
+
|
| 162 |
+
def __init__(
|
| 163 |
+
self,
|
| 164 |
+
W_int: torch.Tensor, # (out_features, in_features) integer codes
|
| 165 |
+
scale: torch.Tensor, # (out_features, num_groups)
|
| 166 |
+
zero: torch.Tensor, # (out_features, num_groups)
|
| 167 |
+
nbits: int,
|
| 168 |
+
group_size: int,
|
| 169 |
+
bias: Optional[torch.Tensor] = None,
|
| 170 |
+
out_dtype: torch.dtype = torch.float16,
|
| 171 |
+
block_n: int = 128,
|
| 172 |
+
num_warps: int = 1,
|
| 173 |
+
) -> None:
|
| 174 |
+
super().__init__()
|
| 175 |
+
if not TRITON_AVAILABLE:
|
| 176 |
+
raise RuntimeError("Triton is not available.")
|
| 177 |
+
if nbits not in SUPPORTED_NBITS:
|
| 178 |
+
raise ValueError(f"Unsupported nbits: {nbits}")
|
| 179 |
+
|
| 180 |
+
out_features, in_features = W_int.shape
|
| 181 |
+
if in_features % group_size != 0:
|
| 182 |
+
raise ValueError("in_features must be divisible by group_size.")
|
| 183 |
+
if group_size % 2 != 0:
|
| 184 |
+
raise ValueError("group_size must be even.")
|
| 185 |
+
|
| 186 |
+
self.out_features = out_features
|
| 187 |
+
self.in_features = in_features
|
| 188 |
+
self.nbits = int(nbits)
|
| 189 |
+
self.group_size = int(group_size)
|
| 190 |
+
self.out_dtype = out_dtype
|
| 191 |
+
self.block_n = int(block_n)
|
| 192 |
+
self.num_warps = int(num_warps)
|
| 193 |
+
self.block_k = self.group_size // 2
|
| 194 |
+
|
| 195 |
+
q_kn = W_int.t().contiguous().to(torch.uint8)
|
| 196 |
+
if nbits == 3:
|
| 197 |
+
self.register_buffer("W_q", _pack_kmajor(q_kn >> 1, 2))
|
| 198 |
+
self.register_buffer("W_lo", _pack_kmajor(q_kn & 1, 1))
|
| 199 |
+
self.eps = 16
|
| 200 |
+
else:
|
| 201 |
+
self.register_buffer("W_q", _pack_kmajor(q_kn, nbits))
|
| 202 |
+
self.register_buffer("W_lo", torch.zeros(1, dtype=torch.int32, device=q_kn.device))
|
| 203 |
+
self.eps = 32 // nbits
|
| 204 |
+
del q_kn
|
| 205 |
+
|
| 206 |
+
self.register_buffer("scale", scale.t().contiguous().to(out_dtype))
|
| 207 |
+
self.register_buffer("zero", zero.t().contiguous().to(out_dtype))
|
| 208 |
+
if bias is not None:
|
| 209 |
+
self.register_buffer("bias", bias.detach().clone().to(out_dtype))
|
| 210 |
+
else:
|
| 211 |
+
self.bias = None
|
| 212 |
+
|
| 213 |
+
# The GEMV accumulates with atomics, so it starts from the bias.
|
| 214 |
+
acc_init = torch.zeros(out_features, dtype=torch.float32, device=self.W_q.device)
|
| 215 |
+
if bias is not None:
|
| 216 |
+
acc_init.copy_(self.bias.float())
|
| 217 |
+
self.register_buffer("_acc_init", acc_init)
|
| 218 |
+
self.register_buffer("_acc", acc_init.clone())
|
| 219 |
+
self._grid = (
|
| 220 |
+
(out_features + self.block_n - 1) // self.block_n,
|
| 221 |
+
in_features // self.group_size,
|
| 222 |
+
)
|
| 223 |
+
|
| 224 |
+
def dequantize_weight(self, dtype: torch.dtype) -> torch.Tensor:
|
| 225 |
+
"""Returns W^T with shape (in_features, out_features)."""
|
| 226 |
+
if self.nbits == 3:
|
| 227 |
+
q = (_unpack_kmajor(self.W_q, 2, self.in_features).to(torch.int32) << 1) | (
|
| 228 |
+
_unpack_kmajor(self.W_lo, 1, self.in_features).to(torch.int32)
|
| 229 |
+
)
|
| 230 |
+
else:
|
| 231 |
+
q = _unpack_kmajor(self.W_q, self.nbits, self.in_features)
|
| 232 |
+
s = self.scale.repeat_interleave(self.group_size, dim=0).to(dtype)
|
| 233 |
+
z = self.zero.repeat_interleave(self.group_size, dim=0).to(dtype)
|
| 234 |
+
return (q.to(dtype) - z) * s
|
| 235 |
+
|
| 236 |
+
# (BLOCK_M, BLOCK_N, SPLIT_K, num_warps, num_stages), largest tile first;
|
| 237 |
+
# the first entry that fits in shared memory is cached per shape.
|
| 238 |
+
_SMALL_M_CONFIGS = ((16, 64, 8, 4, 2), (16, 64, 4, 4, 1))
|
| 239 |
+
_LARGE_M_CONFIGS = ((128, 128, 1, 8, 4), (128, 128, 1, 8, 3),
|
| 240 |
+
(128, 64, 1, 4, 3), (64, 64, 1, 4, 2))
|
| 241 |
+
|
| 242 |
+
def _gemm(self, x2d: torch.Tensor) -> torch.Tensor:
|
| 243 |
+
M = x2d.shape[0]
|
| 244 |
+
N, K, gs = self.out_features, self.in_features, self.group_size
|
| 245 |
+
block_k = min(gs, 32)
|
| 246 |
+
configs = self._SMALL_M_CONFIGS if M <= 32 else self._LARGE_M_CONFIGS
|
| 247 |
+
cache_key = (M <= 32, N, K, gs, self.nbits)
|
| 248 |
+
if cache_key in _GEMM_CONFIG_CACHE:
|
| 249 |
+
configs = (_GEMM_CONFIG_CACHE[cache_key],)
|
| 250 |
+
|
| 251 |
+
tl_dtype = tl.float16 if self.out_dtype == torch.float16 else tl.bfloat16
|
| 252 |
+
last_err = None
|
| 253 |
+
for cfg in configs:
|
| 254 |
+
block_m, block_n, split_k, warps, stages = cfg
|
| 255 |
+
split_k = min(split_k, max(1, K // block_k))
|
| 256 |
+
alloc = torch.empty if split_k == 1 else torch.zeros
|
| 257 |
+
y = alloc(M, N, dtype=self.out_dtype, device=x2d.device)
|
| 258 |
+
grid = (triton.cdiv(M, block_m), triton.cdiv(N, block_n), split_k)
|
| 259 |
+
try:
|
| 260 |
+
_dashq_gemm_kernel[grid](
|
| 261 |
+
x2d, self.W_q, self.W_lo, self.scale, self.zero, y,
|
| 262 |
+
M, N, K,
|
| 263 |
+
self.nbits, self.eps, gs,
|
| 264 |
+
block_m, block_n, block_k, split_k, tl_dtype,
|
| 265 |
+
num_warps=warps, num_stages=stages,
|
| 266 |
+
)
|
| 267 |
+
except triton.runtime.errors.OutOfResources as exc:
|
| 268 |
+
last_err = exc
|
| 269 |
+
continue
|
| 270 |
+
_GEMM_CONFIG_CACHE[cache_key] = cfg
|
| 271 |
+
return y
|
| 272 |
+
raise last_err
|
| 273 |
+
|
| 274 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
| 275 |
+
shape = x.shape
|
| 276 |
+
tokens = x.numel() // shape[-1]
|
| 277 |
+
if tokens == 1 and x.is_cuda:
|
| 278 |
+
self._acc.copy_(self._acc_init)
|
| 279 |
+
_dashq_gemv_kernel[self._grid](
|
| 280 |
+
x.reshape(-1), self.W_q, self.W_lo, self.scale, self.zero, self._acc,
|
| 281 |
+
self.out_features, self.in_features,
|
| 282 |
+
self.nbits, self.eps, self.group_size,
|
| 283 |
+
self.block_n, self.block_k,
|
| 284 |
+
num_warps=self.num_warps,
|
| 285 |
+
)
|
| 286 |
+
return self._acc.to(x.dtype).reshape(*shape[:-1], self.out_features)
|
| 287 |
+
|
| 288 |
+
x2d = x.reshape(tokens, -1)
|
| 289 |
+
if x.is_cuda and TRITON_AVAILABLE:
|
| 290 |
+
out = self._gemm(x2d)
|
| 291 |
+
else:
|
| 292 |
+
out = x2d @ self.dequantize_weight(x.dtype)
|
| 293 |
+
if self.bias is not None:
|
| 294 |
+
out = out + self.bias.to(out.dtype)
|
| 295 |
+
return out.to(x.dtype).reshape(*shape[:-1], self.out_features)
|
| 296 |
+
|
| 297 |
+
def extra_repr(self) -> str:
|
| 298 |
+
return (
|
| 299 |
+
f"in_features={self.in_features}, out_features={self.out_features}, "
|
| 300 |
+
f"nbits={self.nbits}, group_size={self.group_size}"
|
| 301 |
+
)
|
generation_config.json
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"bos_token_id": 100257,
|
| 4 |
+
"eos_token_id": [
|
| 5 |
+
100257,
|
| 6 |
+
100265
|
| 7 |
+
],
|
| 8 |
+
"pad_token_id": 100349,
|
| 9 |
+
"transformers_version": "5.15.0",
|
| 10 |
+
"use_cache": false
|
| 11 |
+
}
|
model-00001-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:86c0d50cbf83fa195504a50944887a3fddea0efcf90cac81b8a2bd1029fd64f2
|
| 3 |
+
size 4962277712
|
model-00002-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:648cd5faac9fa95921604ef2cdf86d4abcbf028863e24cc82dfb35457e41c235
|
| 3 |
+
size 4335553856
|
model.safetensors.index.json
ADDED
|
@@ -0,0 +1,571 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metadata": {
|
| 3 |
+
"total_parameters": 1028019200,
|
| 4 |
+
"total_size": 9297766400
|
| 5 |
+
},
|
| 6 |
+
"weight_map": {
|
| 7 |
+
"lm_head.weight": "model-00002-of-00002.safetensors",
|
| 8 |
+
"model.embed_tokens.weight": "model-00001-of-00002.safetensors",
|
| 9 |
+
"model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 10 |
+
"model.layers.0.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 11 |
+
"model.layers.0.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 12 |
+
"model.layers.0.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 13 |
+
"model.layers.0.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 14 |
+
"model.layers.0.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 15 |
+
"model.layers.0.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 16 |
+
"model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 17 |
+
"model.layers.0.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 18 |
+
"model.layers.0.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 19 |
+
"model.layers.0.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 20 |
+
"model.layers.0.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 21 |
+
"model.layers.0.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 22 |
+
"model.layers.0.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 23 |
+
"model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 24 |
+
"model.layers.1.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 25 |
+
"model.layers.1.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 26 |
+
"model.layers.1.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 27 |
+
"model.layers.1.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 28 |
+
"model.layers.1.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 29 |
+
"model.layers.1.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 30 |
+
"model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 31 |
+
"model.layers.1.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 32 |
+
"model.layers.1.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 33 |
+
"model.layers.1.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 34 |
+
"model.layers.1.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 35 |
+
"model.layers.1.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 36 |
+
"model.layers.1.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 37 |
+
"model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 38 |
+
"model.layers.10.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 39 |
+
"model.layers.10.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 40 |
+
"model.layers.10.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 41 |
+
"model.layers.10.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 42 |
+
"model.layers.10.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 43 |
+
"model.layers.10.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 44 |
+
"model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 45 |
+
"model.layers.10.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 46 |
+
"model.layers.10.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 47 |
+
"model.layers.10.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 48 |
+
"model.layers.10.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 49 |
+
"model.layers.10.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 50 |
+
"model.layers.10.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 51 |
+
"model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 52 |
+
"model.layers.11.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 53 |
+
"model.layers.11.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 54 |
+
"model.layers.11.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 55 |
+
"model.layers.11.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 56 |
+
"model.layers.11.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 57 |
+
"model.layers.11.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 58 |
+
"model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 59 |
+
"model.layers.11.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 60 |
+
"model.layers.11.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 61 |
+
"model.layers.11.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 62 |
+
"model.layers.11.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 63 |
+
"model.layers.11.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 64 |
+
"model.layers.11.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 65 |
+
"model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 66 |
+
"model.layers.12.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 67 |
+
"model.layers.12.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 68 |
+
"model.layers.12.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 69 |
+
"model.layers.12.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 70 |
+
"model.layers.12.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 71 |
+
"model.layers.12.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 72 |
+
"model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 73 |
+
"model.layers.12.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 74 |
+
"model.layers.12.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 75 |
+
"model.layers.12.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 76 |
+
"model.layers.12.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 77 |
+
"model.layers.12.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 78 |
+
"model.layers.12.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 79 |
+
"model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 80 |
+
"model.layers.13.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 81 |
+
"model.layers.13.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 82 |
+
"model.layers.13.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 83 |
+
"model.layers.13.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 84 |
+
"model.layers.13.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 85 |
+
"model.layers.13.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 86 |
+
"model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 87 |
+
"model.layers.13.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 88 |
+
"model.layers.13.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 89 |
+
"model.layers.13.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 90 |
+
"model.layers.13.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 91 |
+
"model.layers.13.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 92 |
+
"model.layers.13.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 93 |
+
"model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 94 |
+
"model.layers.14.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 95 |
+
"model.layers.14.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 96 |
+
"model.layers.14.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 97 |
+
"model.layers.14.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 98 |
+
"model.layers.14.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 99 |
+
"model.layers.14.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 100 |
+
"model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 101 |
+
"model.layers.14.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 102 |
+
"model.layers.14.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 103 |
+
"model.layers.14.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 104 |
+
"model.layers.14.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 105 |
+
"model.layers.14.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 106 |
+
"model.layers.14.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 107 |
+
"model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 108 |
+
"model.layers.15.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 109 |
+
"model.layers.15.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 110 |
+
"model.layers.15.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 111 |
+
"model.layers.15.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 112 |
+
"model.layers.15.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 113 |
+
"model.layers.15.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 114 |
+
"model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 115 |
+
"model.layers.15.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 116 |
+
"model.layers.15.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 117 |
+
"model.layers.15.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 118 |
+
"model.layers.15.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 119 |
+
"model.layers.15.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 120 |
+
"model.layers.15.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 121 |
+
"model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 122 |
+
"model.layers.16.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 123 |
+
"model.layers.16.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 124 |
+
"model.layers.16.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 125 |
+
"model.layers.16.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 126 |
+
"model.layers.16.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 127 |
+
"model.layers.16.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 128 |
+
"model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 129 |
+
"model.layers.16.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 130 |
+
"model.layers.16.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 131 |
+
"model.layers.16.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 132 |
+
"model.layers.16.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 133 |
+
"model.layers.16.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 134 |
+
"model.layers.16.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 135 |
+
"model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 136 |
+
"model.layers.17.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 137 |
+
"model.layers.17.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 138 |
+
"model.layers.17.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 139 |
+
"model.layers.17.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 140 |
+
"model.layers.17.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 141 |
+
"model.layers.17.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 142 |
+
"model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 143 |
+
"model.layers.17.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 144 |
+
"model.layers.17.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 145 |
+
"model.layers.17.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 146 |
+
"model.layers.17.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 147 |
+
"model.layers.17.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 148 |
+
"model.layers.17.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 149 |
+
"model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 150 |
+
"model.layers.18.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 151 |
+
"model.layers.18.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 152 |
+
"model.layers.18.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 153 |
+
"model.layers.18.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 154 |
+
"model.layers.18.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 155 |
+
"model.layers.18.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 156 |
+
"model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 157 |
+
"model.layers.18.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 158 |
+
"model.layers.18.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 159 |
+
"model.layers.18.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 160 |
+
"model.layers.18.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 161 |
+
"model.layers.18.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 162 |
+
"model.layers.18.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 163 |
+
"model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 164 |
+
"model.layers.19.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 165 |
+
"model.layers.19.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 166 |
+
"model.layers.19.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 167 |
+
"model.layers.19.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 168 |
+
"model.layers.19.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 169 |
+
"model.layers.19.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 170 |
+
"model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 171 |
+
"model.layers.19.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 172 |
+
"model.layers.19.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 173 |
+
"model.layers.19.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 174 |
+
"model.layers.19.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 175 |
+
"model.layers.19.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 176 |
+
"model.layers.19.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 177 |
+
"model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 178 |
+
"model.layers.2.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 179 |
+
"model.layers.2.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 180 |
+
"model.layers.2.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 181 |
+
"model.layers.2.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 182 |
+
"model.layers.2.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 183 |
+
"model.layers.2.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 184 |
+
"model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 185 |
+
"model.layers.2.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 186 |
+
"model.layers.2.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 187 |
+
"model.layers.2.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 188 |
+
"model.layers.2.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 189 |
+
"model.layers.2.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 190 |
+
"model.layers.2.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 191 |
+
"model.layers.20.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 192 |
+
"model.layers.20.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 193 |
+
"model.layers.20.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 194 |
+
"model.layers.20.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 195 |
+
"model.layers.20.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 196 |
+
"model.layers.20.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 197 |
+
"model.layers.20.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 198 |
+
"model.layers.20.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 199 |
+
"model.layers.20.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 200 |
+
"model.layers.20.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 201 |
+
"model.layers.20.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 202 |
+
"model.layers.20.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 203 |
+
"model.layers.20.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 204 |
+
"model.layers.20.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 205 |
+
"model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 206 |
+
"model.layers.21.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 207 |
+
"model.layers.21.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 208 |
+
"model.layers.21.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 209 |
+
"model.layers.21.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 210 |
+
"model.layers.21.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 211 |
+
"model.layers.21.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 212 |
+
"model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 213 |
+
"model.layers.21.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 214 |
+
"model.layers.21.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 215 |
+
"model.layers.21.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 216 |
+
"model.layers.21.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 217 |
+
"model.layers.21.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 218 |
+
"model.layers.21.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 219 |
+
"model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 220 |
+
"model.layers.22.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 221 |
+
"model.layers.22.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 222 |
+
"model.layers.22.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 223 |
+
"model.layers.22.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 224 |
+
"model.layers.22.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 225 |
+
"model.layers.22.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 226 |
+
"model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 227 |
+
"model.layers.22.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 228 |
+
"model.layers.22.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 229 |
+
"model.layers.22.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 230 |
+
"model.layers.22.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 231 |
+
"model.layers.22.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 232 |
+
"model.layers.22.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 233 |
+
"model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 234 |
+
"model.layers.23.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 235 |
+
"model.layers.23.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 236 |
+
"model.layers.23.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 237 |
+
"model.layers.23.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 238 |
+
"model.layers.23.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 239 |
+
"model.layers.23.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 240 |
+
"model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 241 |
+
"model.layers.23.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 242 |
+
"model.layers.23.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 243 |
+
"model.layers.23.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 244 |
+
"model.layers.23.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 245 |
+
"model.layers.23.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 246 |
+
"model.layers.23.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 247 |
+
"model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 248 |
+
"model.layers.24.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 249 |
+
"model.layers.24.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 250 |
+
"model.layers.24.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 251 |
+
"model.layers.24.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 252 |
+
"model.layers.24.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 253 |
+
"model.layers.24.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 254 |
+
"model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 255 |
+
"model.layers.24.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 256 |
+
"model.layers.24.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 257 |
+
"model.layers.24.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 258 |
+
"model.layers.24.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 259 |
+
"model.layers.24.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 260 |
+
"model.layers.24.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 261 |
+
"model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 262 |
+
"model.layers.25.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 263 |
+
"model.layers.25.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 264 |
+
"model.layers.25.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 265 |
+
"model.layers.25.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 266 |
+
"model.layers.25.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 267 |
+
"model.layers.25.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 268 |
+
"model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 269 |
+
"model.layers.25.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 270 |
+
"model.layers.25.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 271 |
+
"model.layers.25.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 272 |
+
"model.layers.25.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 273 |
+
"model.layers.25.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 274 |
+
"model.layers.25.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 275 |
+
"model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 276 |
+
"model.layers.26.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 277 |
+
"model.layers.26.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 278 |
+
"model.layers.26.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 279 |
+
"model.layers.26.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 280 |
+
"model.layers.26.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 281 |
+
"model.layers.26.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 282 |
+
"model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 283 |
+
"model.layers.26.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 284 |
+
"model.layers.26.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 285 |
+
"model.layers.26.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 286 |
+
"model.layers.26.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 287 |
+
"model.layers.26.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 288 |
+
"model.layers.26.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 289 |
+
"model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 290 |
+
"model.layers.27.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 291 |
+
"model.layers.27.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 292 |
+
"model.layers.27.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 293 |
+
"model.layers.27.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 294 |
+
"model.layers.27.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 295 |
+
"model.layers.27.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 296 |
+
"model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 297 |
+
"model.layers.27.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 298 |
+
"model.layers.27.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 299 |
+
"model.layers.27.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 300 |
+
"model.layers.27.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 301 |
+
"model.layers.27.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 302 |
+
"model.layers.27.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 303 |
+
"model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 304 |
+
"model.layers.28.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 305 |
+
"model.layers.28.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 306 |
+
"model.layers.28.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 307 |
+
"model.layers.28.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 308 |
+
"model.layers.28.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 309 |
+
"model.layers.28.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 310 |
+
"model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 311 |
+
"model.layers.28.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 312 |
+
"model.layers.28.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 313 |
+
"model.layers.28.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 314 |
+
"model.layers.28.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 315 |
+
"model.layers.28.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 316 |
+
"model.layers.28.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 317 |
+
"model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 318 |
+
"model.layers.29.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 319 |
+
"model.layers.29.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 320 |
+
"model.layers.29.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 321 |
+
"model.layers.29.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 322 |
+
"model.layers.29.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 323 |
+
"model.layers.29.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 324 |
+
"model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 325 |
+
"model.layers.29.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 326 |
+
"model.layers.29.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 327 |
+
"model.layers.29.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 328 |
+
"model.layers.29.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 329 |
+
"model.layers.29.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 330 |
+
"model.layers.29.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 331 |
+
"model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 332 |
+
"model.layers.3.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 333 |
+
"model.layers.3.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 334 |
+
"model.layers.3.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 335 |
+
"model.layers.3.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 336 |
+
"model.layers.3.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 337 |
+
"model.layers.3.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 338 |
+
"model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 339 |
+
"model.layers.3.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 340 |
+
"model.layers.3.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 341 |
+
"model.layers.3.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 342 |
+
"model.layers.3.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 343 |
+
"model.layers.3.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 344 |
+
"model.layers.3.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 345 |
+
"model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 346 |
+
"model.layers.30.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 347 |
+
"model.layers.30.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 348 |
+
"model.layers.30.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 349 |
+
"model.layers.30.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 350 |
+
"model.layers.30.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 351 |
+
"model.layers.30.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 352 |
+
"model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 353 |
+
"model.layers.30.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 354 |
+
"model.layers.30.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 355 |
+
"model.layers.30.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 356 |
+
"model.layers.30.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 357 |
+
"model.layers.30.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 358 |
+
"model.layers.30.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 359 |
+
"model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 360 |
+
"model.layers.31.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 361 |
+
"model.layers.31.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 362 |
+
"model.layers.31.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 363 |
+
"model.layers.31.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 364 |
+
"model.layers.31.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 365 |
+
"model.layers.31.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 366 |
+
"model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 367 |
+
"model.layers.31.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 368 |
+
"model.layers.31.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 369 |
+
"model.layers.31.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 370 |
+
"model.layers.31.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 371 |
+
"model.layers.31.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 372 |
+
"model.layers.31.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 373 |
+
"model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 374 |
+
"model.layers.32.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 375 |
+
"model.layers.32.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 376 |
+
"model.layers.32.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 377 |
+
"model.layers.32.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 378 |
+
"model.layers.32.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 379 |
+
"model.layers.32.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 380 |
+
"model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 381 |
+
"model.layers.32.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 382 |
+
"model.layers.32.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 383 |
+
"model.layers.32.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 384 |
+
"model.layers.32.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 385 |
+
"model.layers.32.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 386 |
+
"model.layers.32.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 387 |
+
"model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 388 |
+
"model.layers.33.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 389 |
+
"model.layers.33.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 390 |
+
"model.layers.33.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 391 |
+
"model.layers.33.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 392 |
+
"model.layers.33.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 393 |
+
"model.layers.33.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 394 |
+
"model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 395 |
+
"model.layers.33.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 396 |
+
"model.layers.33.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 397 |
+
"model.layers.33.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 398 |
+
"model.layers.33.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 399 |
+
"model.layers.33.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 400 |
+
"model.layers.33.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 401 |
+
"model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 402 |
+
"model.layers.34.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 403 |
+
"model.layers.34.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 404 |
+
"model.layers.34.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 405 |
+
"model.layers.34.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 406 |
+
"model.layers.34.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 407 |
+
"model.layers.34.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 408 |
+
"model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 409 |
+
"model.layers.34.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 410 |
+
"model.layers.34.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 411 |
+
"model.layers.34.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 412 |
+
"model.layers.34.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 413 |
+
"model.layers.34.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 414 |
+
"model.layers.34.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 415 |
+
"model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 416 |
+
"model.layers.35.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 417 |
+
"model.layers.35.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 418 |
+
"model.layers.35.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 419 |
+
"model.layers.35.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 420 |
+
"model.layers.35.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 421 |
+
"model.layers.35.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 422 |
+
"model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 423 |
+
"model.layers.35.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 424 |
+
"model.layers.35.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 425 |
+
"model.layers.35.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 426 |
+
"model.layers.35.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 427 |
+
"model.layers.35.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 428 |
+
"model.layers.35.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 429 |
+
"model.layers.36.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 430 |
+
"model.layers.36.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 431 |
+
"model.layers.36.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 432 |
+
"model.layers.36.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 433 |
+
"model.layers.36.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 434 |
+
"model.layers.36.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 435 |
+
"model.layers.36.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 436 |
+
"model.layers.36.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 437 |
+
"model.layers.36.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 438 |
+
"model.layers.36.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 439 |
+
"model.layers.36.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 440 |
+
"model.layers.36.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 441 |
+
"model.layers.36.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 442 |
+
"model.layers.36.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 443 |
+
"model.layers.37.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 444 |
+
"model.layers.37.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 445 |
+
"model.layers.37.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 446 |
+
"model.layers.37.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 447 |
+
"model.layers.37.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 448 |
+
"model.layers.37.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 449 |
+
"model.layers.37.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 450 |
+
"model.layers.37.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 451 |
+
"model.layers.37.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 452 |
+
"model.layers.37.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 453 |
+
"model.layers.37.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 454 |
+
"model.layers.37.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 455 |
+
"model.layers.37.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 456 |
+
"model.layers.37.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 457 |
+
"model.layers.38.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 458 |
+
"model.layers.38.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 459 |
+
"model.layers.38.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 460 |
+
"model.layers.38.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 461 |
+
"model.layers.38.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 462 |
+
"model.layers.38.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 463 |
+
"model.layers.38.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 464 |
+
"model.layers.38.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 465 |
+
"model.layers.38.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 466 |
+
"model.layers.38.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 467 |
+
"model.layers.38.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 468 |
+
"model.layers.38.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 469 |
+
"model.layers.38.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 470 |
+
"model.layers.38.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 471 |
+
"model.layers.39.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 472 |
+
"model.layers.39.mlp.down_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 473 |
+
"model.layers.39.mlp.down_proj.scale": "model-00002-of-00002.safetensors",
|
| 474 |
+
"model.layers.39.mlp.down_proj.zero": "model-00002-of-00002.safetensors",
|
| 475 |
+
"model.layers.39.mlp.gate_up_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 476 |
+
"model.layers.39.mlp.gate_up_proj.scale": "model-00002-of-00002.safetensors",
|
| 477 |
+
"model.layers.39.mlp.gate_up_proj.zero": "model-00002-of-00002.safetensors",
|
| 478 |
+
"model.layers.39.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
| 479 |
+
"model.layers.39.self_attn.o_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 480 |
+
"model.layers.39.self_attn.o_proj.scale": "model-00002-of-00002.safetensors",
|
| 481 |
+
"model.layers.39.self_attn.o_proj.zero": "model-00002-of-00002.safetensors",
|
| 482 |
+
"model.layers.39.self_attn.qkv_proj.W_q_packed": "model-00002-of-00002.safetensors",
|
| 483 |
+
"model.layers.39.self_attn.qkv_proj.scale": "model-00002-of-00002.safetensors",
|
| 484 |
+
"model.layers.39.self_attn.qkv_proj.zero": "model-00002-of-00002.safetensors",
|
| 485 |
+
"model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 486 |
+
"model.layers.4.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 487 |
+
"model.layers.4.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 488 |
+
"model.layers.4.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 489 |
+
"model.layers.4.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 490 |
+
"model.layers.4.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 491 |
+
"model.layers.4.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 492 |
+
"model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 493 |
+
"model.layers.4.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 494 |
+
"model.layers.4.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 495 |
+
"model.layers.4.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 496 |
+
"model.layers.4.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 497 |
+
"model.layers.4.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 498 |
+
"model.layers.4.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 499 |
+
"model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 500 |
+
"model.layers.5.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 501 |
+
"model.layers.5.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 502 |
+
"model.layers.5.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 503 |
+
"model.layers.5.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 504 |
+
"model.layers.5.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 505 |
+
"model.layers.5.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 506 |
+
"model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 507 |
+
"model.layers.5.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 508 |
+
"model.layers.5.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 509 |
+
"model.layers.5.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 510 |
+
"model.layers.5.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 511 |
+
"model.layers.5.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 512 |
+
"model.layers.5.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 513 |
+
"model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 514 |
+
"model.layers.6.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 515 |
+
"model.layers.6.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 516 |
+
"model.layers.6.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 517 |
+
"model.layers.6.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 518 |
+
"model.layers.6.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 519 |
+
"model.layers.6.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 520 |
+
"model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 521 |
+
"model.layers.6.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 522 |
+
"model.layers.6.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 523 |
+
"model.layers.6.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 524 |
+
"model.layers.6.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 525 |
+
"model.layers.6.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 526 |
+
"model.layers.6.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 527 |
+
"model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 528 |
+
"model.layers.7.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 529 |
+
"model.layers.7.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 530 |
+
"model.layers.7.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 531 |
+
"model.layers.7.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 532 |
+
"model.layers.7.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 533 |
+
"model.layers.7.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 534 |
+
"model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 535 |
+
"model.layers.7.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 536 |
+
"model.layers.7.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 537 |
+
"model.layers.7.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 538 |
+
"model.layers.7.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 539 |
+
"model.layers.7.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 540 |
+
"model.layers.7.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 541 |
+
"model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 542 |
+
"model.layers.8.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 543 |
+
"model.layers.8.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 544 |
+
"model.layers.8.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 545 |
+
"model.layers.8.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 546 |
+
"model.layers.8.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 547 |
+
"model.layers.8.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 548 |
+
"model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 549 |
+
"model.layers.8.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 550 |
+
"model.layers.8.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 551 |
+
"model.layers.8.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 552 |
+
"model.layers.8.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 553 |
+
"model.layers.8.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 554 |
+
"model.layers.8.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 555 |
+
"model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 556 |
+
"model.layers.9.mlp.down_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 557 |
+
"model.layers.9.mlp.down_proj.scale": "model-00001-of-00002.safetensors",
|
| 558 |
+
"model.layers.9.mlp.down_proj.zero": "model-00001-of-00002.safetensors",
|
| 559 |
+
"model.layers.9.mlp.gate_up_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 560 |
+
"model.layers.9.mlp.gate_up_proj.scale": "model-00001-of-00002.safetensors",
|
| 561 |
+
"model.layers.9.mlp.gate_up_proj.zero": "model-00001-of-00002.safetensors",
|
| 562 |
+
"model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 563 |
+
"model.layers.9.self_attn.o_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 564 |
+
"model.layers.9.self_attn.o_proj.scale": "model-00001-of-00002.safetensors",
|
| 565 |
+
"model.layers.9.self_attn.o_proj.zero": "model-00001-of-00002.safetensors",
|
| 566 |
+
"model.layers.9.self_attn.qkv_proj.W_q_packed": "model-00001-of-00002.safetensors",
|
| 567 |
+
"model.layers.9.self_attn.qkv_proj.scale": "model-00001-of-00002.safetensors",
|
| 568 |
+
"model.layers.9.self_attn.qkv_proj.zero": "model-00001-of-00002.safetensors",
|
| 569 |
+
"model.norm.weight": "model-00002-of-00002.safetensors"
|
| 570 |
+
}
|
| 571 |
+
}
|
modeling_dashq.py
ADDED
|
@@ -0,0 +1,227 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Inference code for this DASH-Q checkpoint.
|
| 2 |
+
|
| 3 |
+
Generated by export_hf_repo.py -- do not edit by hand.
|
| 4 |
+
|
| 5 |
+
Weights are group-wise asymmetric integers packed into int32 words; the layout of
|
| 6 |
+
each quantized layer is described by `dashq_config.json`. At load time the layers
|
| 7 |
+
are converted to the format used by the Triton kernels in `dashq_kernel.py`, with
|
| 8 |
+
a PyTorch dequantize-and-matmul fallback when Triton is unavailable.
|
| 9 |
+
"""
|
| 10 |
+
from __future__ import annotations
|
| 11 |
+
|
| 12 |
+
import json
|
| 13 |
+
import os
|
| 14 |
+
from typing import Any, Dict, Optional
|
| 15 |
+
|
| 16 |
+
import torch
|
| 17 |
+
import torch.nn as nn
|
| 18 |
+
import torch.nn.functional as F
|
| 19 |
+
|
| 20 |
+
from transformers import AutoConfig, Phi3ForCausalLM
|
| 21 |
+
|
| 22 |
+
try:
|
| 23 |
+
from .dashq_kernel import TRITON_AVAILABLE, SUPPORTED_NBITS, TritonQuantLinear
|
| 24 |
+
except ImportError: # loaded as a flat module by trust_remote_code
|
| 25 |
+
from dashq_kernel import TRITON_AVAILABLE, SUPPORTED_NBITS, TritonQuantLinear
|
| 26 |
+
|
| 27 |
+
DASHQ_CONFIG_FILE = "dashq_config.json"
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
def _unpack_int_values(packed: torch.Tensor, nbits: int, numel: int) -> torch.Tensor:
|
| 31 |
+
values_per_word = max(1, 32 // nbits)
|
| 32 |
+
mask = (1 << nbits) - 1
|
| 33 |
+
shifts = torch.arange(values_per_word, device=packed.device, dtype=torch.int32) * nbits
|
| 34 |
+
out = (packed.view(-1, 1) >> shifts.view(1, -1)) & mask
|
| 35 |
+
return out.reshape(-1)[:numel]
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
class DashQPackedLinear(nn.Module):
|
| 39 |
+
"""Checkpoint buffers for one quantized layer."""
|
| 40 |
+
|
| 41 |
+
def __init__(self, in_features: int, out_features: int, nbits: int, group_size: int,
|
| 42 |
+
bias: bool, dtype: torch.dtype, quant_in_features: Optional[int] = None) -> None:
|
| 43 |
+
super().__init__()
|
| 44 |
+
self.in_features = int(in_features)
|
| 45 |
+
self.quant_in_features = int(quant_in_features or in_features)
|
| 46 |
+
self.out_features = int(out_features)
|
| 47 |
+
self.nbits = int(nbits)
|
| 48 |
+
self.group_size = int(group_size)
|
| 49 |
+
self.linear_dtype = dtype
|
| 50 |
+
self.numel = self.out_features * self.quant_in_features
|
| 51 |
+
self.num_groups = self.numel // self.group_size
|
| 52 |
+
values_per_word = max(1, 32 // self.nbits)
|
| 53 |
+
n_words = (self.numel + values_per_word - 1) // values_per_word
|
| 54 |
+
|
| 55 |
+
self.register_buffer("W_q_packed", torch.zeros(n_words, dtype=torch.int32))
|
| 56 |
+
self.register_buffer("scale", torch.zeros(self.num_groups, 1, dtype=torch.float16))
|
| 57 |
+
self.register_buffer("zero", torch.zeros(self.num_groups, 1, dtype=torch.float16))
|
| 58 |
+
if bias:
|
| 59 |
+
self.bias = nn.Parameter(torch.zeros(self.out_features, dtype=dtype), requires_grad=False)
|
| 60 |
+
else:
|
| 61 |
+
self.register_parameter("bias", None)
|
| 62 |
+
self.kernel: Optional[nn.Module] = None
|
| 63 |
+
|
| 64 |
+
def dequantize_weight(self, dtype: torch.dtype) -> torch.Tensor:
|
| 65 |
+
W_int = _unpack_int_values(self.W_q_packed, self.nbits, self.numel)
|
| 66 |
+
W_int = W_int.view(self.num_groups, self.group_size).to(dtype)
|
| 67 |
+
W = (W_int - self.zero.to(dtype)) * self.scale.to(dtype)
|
| 68 |
+
return W.view(self.out_features, self.quant_in_features)
|
| 69 |
+
|
| 70 |
+
@torch.no_grad()
|
| 71 |
+
def build_kernel(self) -> bool:
|
| 72 |
+
if self.kernel is not None:
|
| 73 |
+
return True
|
| 74 |
+
if not TRITON_AVAILABLE or self.nbits not in SUPPORTED_NBITS:
|
| 75 |
+
return False
|
| 76 |
+
if self.quant_in_features % self.group_size or self.group_size % 2:
|
| 77 |
+
return False
|
| 78 |
+
if self.W_q_packed is None or not self.W_q_packed.is_cuda:
|
| 79 |
+
return False
|
| 80 |
+
W_int = _unpack_int_values(self.W_q_packed, self.nbits, self.numel)
|
| 81 |
+
W_int = W_int.view(self.out_features, self.quant_in_features)
|
| 82 |
+
ng = self.quant_in_features // self.group_size
|
| 83 |
+
self.kernel = TritonQuantLinear(
|
| 84 |
+
W_int,
|
| 85 |
+
self.scale.view(self.out_features, ng),
|
| 86 |
+
self.zero.view(self.out_features, ng),
|
| 87 |
+
self.nbits,
|
| 88 |
+
self.group_size,
|
| 89 |
+
bias=self.bias.data if self.bias is not None else None,
|
| 90 |
+
out_dtype=self.linear_dtype,
|
| 91 |
+
)
|
| 92 |
+
del W_int
|
| 93 |
+
self._buffers["W_q_packed"] = None
|
| 94 |
+
self._buffers["scale"] = None
|
| 95 |
+
self._buffers["zero"] = None
|
| 96 |
+
return True
|
| 97 |
+
|
| 98 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
| 99 |
+
if self.kernel is not None:
|
| 100 |
+
return self.kernel(x)
|
| 101 |
+
weight = self.dequantize_weight(x.dtype)
|
| 102 |
+
bias = self.bias.to(x.dtype) if self.bias is not None else None
|
| 103 |
+
return F.linear(x, weight, bias)
|
| 104 |
+
|
| 105 |
+
def extra_repr(self) -> str:
|
| 106 |
+
return (f"in_features={self.in_features}, out_features={self.out_features}, "
|
| 107 |
+
f"nbits={self.nbits}, group_size={self.group_size}")
|
| 108 |
+
|
| 109 |
+
|
| 110 |
+
def _set_module(root: nn.Module, name: str, new_module: nn.Module) -> None:
|
| 111 |
+
parts = name.split(".")
|
| 112 |
+
parent = root
|
| 113 |
+
for part in parts[:-1]:
|
| 114 |
+
parent = getattr(parent, part)
|
| 115 |
+
setattr(parent, parts[-1], new_module)
|
| 116 |
+
|
| 117 |
+
|
| 118 |
+
def _get_module(root: nn.Module, name: str) -> Optional[nn.Module]:
|
| 119 |
+
obj = root
|
| 120 |
+
for part in name.split("."):
|
| 121 |
+
if not hasattr(obj, part):
|
| 122 |
+
return None
|
| 123 |
+
obj = getattr(obj, part)
|
| 124 |
+
return obj
|
| 125 |
+
|
| 126 |
+
|
| 127 |
+
_DTYPES = {"float16": torch.float16, "bfloat16": torch.bfloat16, "float32": torch.float32}
|
| 128 |
+
|
| 129 |
+
|
| 130 |
+
def _load_dashq_spec(model_id_or_path, **kwargs) -> Dict[str, Any]:
|
| 131 |
+
"""Read dashq_config.json from a local dir or the Hub."""
|
| 132 |
+
path = None
|
| 133 |
+
if model_id_or_path is not None:
|
| 134 |
+
local = os.path.join(str(model_id_or_path), DASHQ_CONFIG_FILE)
|
| 135 |
+
if os.path.isfile(local):
|
| 136 |
+
path = local
|
| 137 |
+
if path is None and model_id_or_path is not None:
|
| 138 |
+
try:
|
| 139 |
+
from huggingface_hub import hf_hub_download
|
| 140 |
+
|
| 141 |
+
path = hf_hub_download(
|
| 142 |
+
repo_id=str(model_id_or_path),
|
| 143 |
+
filename=DASHQ_CONFIG_FILE,
|
| 144 |
+
revision=kwargs.get("revision"),
|
| 145 |
+
token=kwargs.get("token"),
|
| 146 |
+
cache_dir=kwargs.get("cache_dir"),
|
| 147 |
+
)
|
| 148 |
+
except Exception:
|
| 149 |
+
return {}
|
| 150 |
+
if path is None:
|
| 151 |
+
return {}
|
| 152 |
+
with open(path, encoding="utf-8") as f:
|
| 153 |
+
return json.load(f)
|
| 154 |
+
|
| 155 |
+
|
| 156 |
+
def _swap_quantized_modules(model: nn.Module, modules: Dict[str, Any]) -> int:
|
| 157 |
+
count = 0
|
| 158 |
+
for name, meta in modules.items():
|
| 159 |
+
target = _get_module(model, name)
|
| 160 |
+
if target is None or isinstance(target, DashQPackedLinear):
|
| 161 |
+
continue
|
| 162 |
+
module = DashQPackedLinear(
|
| 163 |
+
in_features=meta["in_features"],
|
| 164 |
+
out_features=meta["out_features"],
|
| 165 |
+
nbits=meta["nbits"],
|
| 166 |
+
group_size=meta["group_size"],
|
| 167 |
+
bias=getattr(target, "bias", None) is not None,
|
| 168 |
+
dtype=_DTYPES.get(meta.get("linear_dtype", "float16"), torch.float16),
|
| 169 |
+
quant_in_features=meta.get("quant_in_features"),
|
| 170 |
+
)
|
| 171 |
+
_set_module(model, name, module)
|
| 172 |
+
count += 1
|
| 173 |
+
return count
|
| 174 |
+
|
| 175 |
+
|
| 176 |
+
class DashQPhi3ForCausalLM(Phi3ForCausalLM):
|
| 177 |
+
"""Phi3ForCausalLM whose linear layers hold DASH-Q packed quantized weights."""
|
| 178 |
+
|
| 179 |
+
def __init__(self, config):
|
| 180 |
+
super().__init__(config)
|
| 181 |
+
modules = (getattr(config, "dashq_modules", None) or {})
|
| 182 |
+
if modules:
|
| 183 |
+
_swap_quantized_modules(self, modules)
|
| 184 |
+
|
| 185 |
+
@classmethod
|
| 186 |
+
def from_pretrained(cls, pretrained_model_name_or_path=None, *args, **kwargs):
|
| 187 |
+
config = kwargs.pop("config", None)
|
| 188 |
+
if config is None and pretrained_model_name_or_path is not None:
|
| 189 |
+
config = AutoConfig.from_pretrained(
|
| 190 |
+
pretrained_model_name_or_path,
|
| 191 |
+
trust_remote_code=True,
|
| 192 |
+
revision=kwargs.get("revision"),
|
| 193 |
+
token=kwargs.get("token"),
|
| 194 |
+
cache_dir=kwargs.get("cache_dir"),
|
| 195 |
+
)
|
| 196 |
+
if config is not None and not getattr(config, "dashq_modules", None):
|
| 197 |
+
spec = _load_dashq_spec(pretrained_model_name_or_path, **kwargs)
|
| 198 |
+
config.dashq_modules = spec.get("quantized_modules", {})
|
| 199 |
+
model = super().from_pretrained(pretrained_model_name_or_path, *args, config=config, **kwargs)
|
| 200 |
+
model.build_dashq_kernels()
|
| 201 |
+
return model
|
| 202 |
+
|
| 203 |
+
def build_dashq_kernels(self, verbose: bool = True) -> "DashQPhi3ForCausalLM":
|
| 204 |
+
"""Move the packed buffers to the Triton kernel layout (no-op off CUDA)."""
|
| 205 |
+
total = built = 0
|
| 206 |
+
for module in self.modules():
|
| 207 |
+
if isinstance(module, DashQPackedLinear):
|
| 208 |
+
total += 1
|
| 209 |
+
built += int(module.build_kernel())
|
| 210 |
+
if verbose and total:
|
| 211 |
+
if built:
|
| 212 |
+
print(f">> DASH-Q: {built}/{total} linear layers using the Triton decode kernel.")
|
| 213 |
+
else:
|
| 214 |
+
print(f">> DASH-Q: {total} quantized layers using the PyTorch fallback path.")
|
| 215 |
+
return self
|
| 216 |
+
|
| 217 |
+
def save_pretrained(self, *args, **kwargs):
|
| 218 |
+
released = any(
|
| 219 |
+
isinstance(m, DashQPackedLinear) and m.kernel is not None for m in self.modules()
|
| 220 |
+
)
|
| 221 |
+
if released:
|
| 222 |
+
raise RuntimeError(
|
| 223 |
+
"This model has been converted to the DASH-Q Triton kernel, so the packed "
|
| 224 |
+
"buffers are no longer materialized and saving would produce an incomplete "
|
| 225 |
+
"checkpoint. Reload with build_dashq_kernels() skipped if you need to re-save."
|
| 226 |
+
)
|
| 227 |
+
return super().save_pretrained(*args, **kwargs)
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|im_end|>",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 16384,
|
| 10 |
+
"pad_token": "<|dummy_85|>",
|
| 11 |
+
"tokenizer_class": "TokenizersBackend"
|
| 12 |
+
}
|