File size: 5,653 Bytes
47f6973 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 |
---
language:
- en
pipeline_tag: text-generation
license: apache-2.0
---
# SmolLM-135M-Instruct-quantized.w4a16
## Model Overview
- **Model Architecture:** SmolLM-135M-Instruct
- **Input:** Text
- **Output:** Text
- **Model Optimizations:**
- **Weight quantization:** INT4
- **Intended Use Cases:** Intended for commercial and research use in English. Similarly to [SmolLM-135M-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM-135M), this models is intended for assistant-like chat.
- **Out-of-scope:** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English.
- **Release Date:** 8/23/2024
- **Version:** 1.0
- **License(s)**: [Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)
- **Model Developers:** Neural Magic
Quantized version of [SmolLM-135M-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM-135M).
It achieves an average score of 31.91 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 31.55.
### Model Optimizations
This model was obtained by quantizing the weights of [SmolLM-135M-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM-135M) to INT4 data type.
This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%.
Only the weights of the linear operators within transformers blocks are quantized. Symmetric group-wise quantization is applied, in which a linear scaling per group maps the INT4 and floating point representations of the quantized weights.
The [GPTQ](https://arxiv.org/abs/2210.17323) algorithm is applied for quantization, as implemented in the [llm-compressor](https://github.com/vllm-project/llm-compressor) library. Quantization is performed with 10% damping factor, group-size as 64 and 512 sequences sampled from [LLM Compression Calibration](https://huggingface.co/datasets/neuralmagic/LLM_compression_calibration).
## Creation
This model was created by using the [llm-compressor](https://github.com/vllm-project/llm-compressor) library as presented in the code snipet below.
```python
from transformers import AutoTokenizer
from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot
from llmcompressor.modifiers.quantization import GPTQModifier
from compressed_tensors.quantization import QuantizationArgs, QuantizationType, QuantizationStrategy
from datasets import load_dataset
import random
model_id = "HuggingFaceTB/SmolLM-135M-Instruct"
num_samples = 512
max_seq_len = 4096
tokenizer = AutoTokenizer.from_pretrained(model_id)
preprocess_fn = lambda example: {"text": "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n{text}".format_map(example)}
dataset_name = "neuralmagic/LLM_compression_calibration"
dataset = load_dataset(dataset_name, split="train")
ds = dataset.shuffle().select(range(num_samples))
ds = ds.map(preprocess_fn)
examples = [
tokenizer(
example["text"], padding=False, max_length=max_seq_len, truncation=True,
) for example in ds
]
# recipe = "w4a16_nohead_recipe.yaml"
recipe = GPTQModifier(
targets="Linear",
scheme="W4A16",
ignore=["lm_head"],
dampening_frac=0.1,
)
model = SparseAutoModelForCausalLM.from_pretrained(
model_id,
device_map="auto",
trust_remote_code=True
)
print(model)
oneshot(
model=model,
dataset=ds,
recipe=recipe,
max_seq_length=max_seq_len,
num_calibration_samples=num_samples,
oneshot_device="cuda:1,2,3",
)
model_name = model_id.split("/")[-1]
model.save_pretrained(f"{model_name}-quantized.w4a16")
tokenizer.save_pretrained(f"{model_name}-quantized.w4a16")
```
## Evaluation
The model was evaluated on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) leaderboard tasks (version 1) with the [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness/tree/383bbd54bc621086e05aa1b030d8d4d5635b25e6) (commit 383bbd54bc621086e05aa1b030d8d4d5635b25e6) and the [sparseML](https://github.com/neuralmagic/sparseml) engine, using the following command:
```
lm_eval \
--model sparseml \
--model_args pretrained=nm-testing/SmolLM-1.7B-Instruct-quantized.w4a16,dtype=bfloat16,max_legth=2048,add_bos_token=True,parallelize=True \
--tasks openllm \
--batch_size auto
```
### Accuracy
#### Open LLM Leaderboard evaluation scores
<table>
<tr>
<td><strong>Benchmark</strong>
</td>
<td><strong>SmolLM-135M-Instruct </strong>
</td>
<td><strong>SmolLM-135M-Instruct-quantized.w4a16(this model)</strong>
</td>
<td><strong>Recovery</strong>
</td>
</tr>
<tr>
<td>MMLU (5-shot)
</td>
<td>26.220
</td>
<td>25.202
</td>
<td>96.12%
</td>
</tr>
<tr>
<td>ARC Challenge (25-shot)
</td>
<td>29.948
</td>
<td>30.034
</td>
<td>100.29%
</td>
</tr>
<tr>
<td>GSM-8K (5-shot, strict-match)
</td>
<td>1.289
</td>
<td>1.971
</td>
<td>152.91%
</td>
</tr>
<tr>
<td>Hellaswag (10-shot)
</td>
<td>41.41
</td>
<td>40.81
</td>
<td>98.55%
</td>
</tr>
<tr>
<td>Winogrande (5-shot)
</td>
<td>50.039
</td>
<td>53.591
</td>
<td>107.10%
</td>
</tr>
<tr>
<td>TruthfulQA (0-shot)
</td>
<td>40.38
</td>
<td>39.87
</td>
<td>98.74%
</td>
</tr>
<tr>
<td><strong>Average</strong>
</td>
<td><strong>31.55</strong>
</td>
<td><strong>31.91</strong>
</td>
<td><strong>101.16%</strong>
</td>
</tr>
</table> |