File size: 10,242 Bytes
d53b1b3 6043ef6 d53b1b3 c8b4ee2 d53b1b3 b758978 d53b1b3 b758978 492f5db b758978 d53b1b3 b758978 d53b1b3 6043ef6 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 |
---
tags:
- fp8
- vllm
license: apache-2.0
license_link: https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md
language:
- en
base_model: ibm-granite/granite-3.1-2b-instruct
library_name: transformers
---
# granite-3.1-2b-instruct-FP8-dynamic
## Model Overview
- **Model Architecture:** granite-3.1-2b-instruct
- **Input:** Text
- **Output:** Text
- **Model Optimizations:**
- **Weight quantization:** FP8
- **Activation quantization:** FP8
- **Release Date:** 1/8/2025
- **Version:** 1.0
- **Model Developers:** Neural Magic
Quantized version of [ibm-granite/granite-3.1-2b-instruct](https://huggingface.co/ibm-granite/granite-3.1-2b-instruct).
It achieves an average score of 61.84 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 61.98.
### Model Optimizations
This model was obtained by quantizing the weights and activations of [ibm-granite/granite-3.1-2b-instruct](https://huggingface.co/ibm-granite/granite-3.1-2b-instruct) to FP8 data type, ready for inference with vLLM >= 0.5.2.
This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%. Only the weights and activations of the linear operators within transformers blocks are quantized.
## Deployment
### Use with vLLM
This model can be deployed efficiently using the [vLLM](https://docs.vllm.ai/en/latest/) backend, as shown in the example below.
```python
from transformers import AutoTokenizer
from vllm import LLM, SamplingParams
max_model_len, tp_size = 4096, 1
model_name = "neuralmagic-ent/granite-3.1-2b-instruct-FP8-dynamic"
tokenizer = AutoTokenizer.from_pretrained(model_name)
llm = LLM(model=model_name, tensor_parallel_size=tp_size, max_model_len=max_model_len, trust_remote_code=True)
sampling_params = SamplingParams(temperature=0.3, max_tokens=256, stop_token_ids=[tokenizer.eos_token_id])
messages_list = [
[{"role": "user", "content": "Who are you? Please respond in pirate speak!"}],
]
prompt_token_ids = [tokenizer.apply_chat_template(messages, add_generation_prompt=True) for messages in messages_list]
outputs = llm.generate(prompt_token_ids=prompt_token_ids, sampling_params=sampling_params)
generated_text = [output.outputs[0].text for output in outputs]
print(generated_text)
```
vLLM also supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details.
## Creation
This model was created with [llm-compressor](https://github.com/vllm-project/llm-compressor) by running the code snippet below.
```bash
python quantize.py --model_id ibm-granite/granite-3.1-2b-base --save_path "output_dir/"
```
```python
import argparse
from transformers import AutoModelForCausalLM, AutoTokenizer
from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.transformers import oneshot
import os
def main():
parser = argparse.ArgumentParser(description='Quantize a transformer model to FP8')
parser.add_argument('--model_id', type=str, required=True,
help='The model ID from HuggingFace (e.g., "meta-llama/Meta-Llama-3-2b-Instruct")')
parser.add_argument('--save_path', type=str, default='.',
help='Custom path to save the quantized model. If not provided, will use model_name-FP8-dynamic')
args = parser.parse_args()
# Load model
model = AutoModelForCausalLM.from_pretrained(
args.model_id, device_map="auto", torch_dtype="auto", trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(args.model_id)
# Configure the quantization algorithm and scheme
recipe = QuantizationModifier(
targets="Linear", scheme="FP8_DYNAMIC", ignore=["lm_head"]
)
# Apply quantization
oneshot(model=model, recipe=recipe)
save_path = os.path.join(args.save_path, args.model_id.split("/")[1] + "-FP8-dynamic")
os.makedirs(save_path, exist_ok=True)
# Save to disk in compressed-tensors format
model.save_pretrained(save_path)
tokenizer.save_pretrained(save_path)
print(f"Model and tokenizer saved to: {save_path}")
if __name__ == "__main__":
main()
```
## Evaluation
The model was evaluated on OpenLLM Leaderboard [V1](https://huggingface.co/spaces/open-llm-leaderboard-old/open_llm_leaderboard) and on [HumanEval](https://github.com/neuralmagic/evalplus), using the following commands:
OpenLLM Leaderboard V1:
```
lm_eval \
--model vllm \
--model_args pretrained="neuralmagic-ent/granite-3.1-2b-instruct-FP8-dynamic",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=1,gpu_memory_utilization=0.8,enable_chunked_prefill=True,trust_remote_code=True \
--tasks openllm \
--write_out \
--batch_size auto \
--output_path output_dir \
--show_config
```
#### HumanEval
##### Generation
```
python3 codegen/generate.py \
--model neuralmagic-ent/granite-3.1-2b-instruct-FP8-dynamic \
--bs 16 \
--temperature 0.2 \
--n_samples 50 \
--root "." \
--dataset humaneval
```
##### Sanitization
```
python3 evalplus/sanitize.py \
humaneval/neuralmagic-ent--granite-3.1-2b-instruct-FP8-dynamic_vllm_temp_0.2
```
##### Evaluation
```
evalplus.evaluate \
--dataset humaneval \
--samples humaneval/neuralmagic-ent--granite-3.1-2b-instruct-FP8-dynamic_vllm_temp_0.2-sanitized
```
### Accuracy
#### OpenLLM Leaderboard V1 evaluation scores
| Metric | ibm-granite/granite-3.1-2b-instruct | neuralmagic-ent/granite-3.1-2b-instruct-FP8-dynamic |
|-----------------------------------------|:---------------------------------:|:-------------------------------------------:|
| ARC-Challenge (Acc-Norm, 25-shot) | 55.63 | 55.03 |
| GSM8K (Strict-Match, 5-shot) | 60.96 | 61.49 |
| HellaSwag (Acc-Norm, 10-shot) | 75.21 | 75.26 |
| MMLU (Acc, 5-shot) | 54.38 | 54.24 |
| TruthfulQA (MC2, 0-shot) | 55.93 | 55.42 |
| Winogrande (Acc, 5-shot) | 69.67 | 69.61 |
| **Average Score** | **61.98** | **61.84** |
| **Recovery** | **100.00** | **99.78** |
| Metric | ibm-granite/granite-3.1-2b-instruct | neuralmagic-ent/granite-3.1-2b-instruct-FP8-dynamic |
|-----------------------------------------|:---------------------------------:|:-------------------------------------------:|
| IFEval (Inst Level Strict Acc, 0-shot)| 67.99 | 66.79 |
| BBH (Acc-Norm, 3-shot) | 44.11 | 44.24 |
| Math-Hard (Exact-Match, 4-shot) | 8.66 | 7.89 |
| GPQA (Acc-Norm, 0-shot) | 28.30 | 26.90 |
| MUSR (Acc-Norm, 0-shot) | 35.12 | 35.12 |
| MMLU-Pro (Acc, 5-shot) | 26.87 | 28.33 |
| **Average Score** | **35.17** | **34.88** |
| **Recovery** | **100.00** | **99.16** |
#### HumanEval pass@1 scores
| Metric | ibm-granite/granite-3.1-2b-instruct | neuralmagic-ent/granite-3.1-2b-instruct-FP8-dynamic |
|-----------------------------------------|:---------------------------------:|:-------------------------------------------:|
| HumanEval Pass@1 | 53.40 | 54.90 |
## Inference Performance
This model achieves up to 1.2x speedup in single-stream deployment on L40 GPUs.
The following performance benchmarks were conducted with [vLLM](https://docs.vllm.ai/en/latest/) version 0.6.6.post1, and [GuideLLM](https://github.com/neuralmagic/guidellm).
### Single-stream performance (measured with vLLM version 0.6.6.post1)
<table>
<tr>
<td></td>
<td></td>
<td></td>
<th style="text-align: center;" colspan="7" >Latency (s)</th>
</tr>
<tr>
<th>GPU class</th>
<th>Model</th>
<th>Speedup</th>
<th>Code Completion<br>prefill: 256 tokens<br>decode: 1024 tokens</th>
<th>Docstring Generation<br>prefill: 768 tokens<br>decode: 128 tokens</th>
<th>Code Fixing<br>prefill: 1024 tokens<br>decode: 1024 tokens</th>
<th>RAG<br>prefill: 1024 tokens<br>decode: 128 tokens</th>
<th>Instruction Following<br>prefill: 256 tokens<br>decode: 128 tokens</th>
<th>Multi-turn Chat<br>prefill: 512 tokens<br>decode: 256 tokens</th>
<th>Large Summarization<br>prefill: 4096 tokens<br>decode: 512 tokens</th>
</tr>
<tr>
<td style="vertical-align: middle;" rowspan="3" >L40</td>
<td>granite-3.1-2b-instruct</td>
<td></td>
<td>9.3</td>
<td>1.2</td>
<td>9.4</td>
<td>1.2</td>
<td>1.2</td>
<td>2.3</td>
<td>5.0</td>
</tr>
<tr>
<td>granite-3.1-2b-instruct-FP8-dynamic<br>(this model)</td>
<td>1.26</td>
<td>7.3</td>
<td>0.9</td>
<td>7.4</td>
<td>1.0</td>
<td>0.9</td>
<td>1.8</td>
<td>4.1</td>
</tr>
<tr>
<td>granite-3.1-2b-instruct-quantized.w4a16</td>
<td>1.88</td>
<td>4.8</td>
<td>0.6</td>
<td>4.9</td>
<td>0.6</td>
<td>0.6</td>
<td>1.2</td>
<td>2.8</td>
</tr>
</table> |