Qwen2.5-3B-quantized.w4a16

Qwen2.5-3B-quantized.w4a16

Model Overview

• Model Architecture: Qwen2
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: INT4
• Intended Use Cases: Intended for commercial and research use multiple languages. Similarly to Qwen2.5-3B, 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).
• Release Date: 12/17/2024
• Version: 1.0
• Model Developers: Neural Magic

Quantized version of Qwen2.5-3B. It achieves an average score of 62.18 on the OpenLLM benchmark (version 1), whereas the unquantized model achieves 63.59.

Model Optimizations

This model was obtained by quantizing the weights and activations of Qwen2.5-3B to INT8 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 weights of the linear operators within transformers blocks are quantized. Symmetric per-group quantization is applied, in which a linear scaling per group of 64 parameters maps the INT4 and floating point representations of the quantized weights.

Deployment

This model can be deployed efficiently using the vLLM backend, as shown in the example below.

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = "neuralmagic-ent/Qwen2.5-3B-quantized.w4a16"
number_gpus = 1
max_model_len = 8192

sampling_params = SamplingParams(temperature=0.7, top_p=0.8, max_tokens=256)

tokenizer = AutoTokenizer.from_pretrained(model_id)

prompt = "Give me a short introduction to large language model."

llm = LLM(model=model_id, tensor_parallel_size=number_gpus, max_model_len=max_model_len)

outputs = llm.generate(prompt, sampling_params)

generated_text = outputs[0].outputs[0].text
print(generated_text)

vLLM aslo supports OpenAI-compatible serving. See the documentation for more details.

Evaluation

The model was evaluated on the OpenLLM leaderboard tasks (version 1) with the lm-evaluation-harness (commit 383bbd54bc621086e05aa1b030d8d4d5635b25e6) and the vLLM engine, using the following command:

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic-ent/Qwen2.5-3B-quantized.w4a16",dtype=auto,gpu_memory_utilization=0.9,add_bos_token=True,max_model_len=4096,enable_chunk_prefill=True,tensor_parallel_size=1 \
  --tasks openllm \
  --batch_size auto

Accuracy

Open LLM Leaderboard evaluation scores

<table> <tr> <td><strong>Benchmark</strong> </td> <td><strong>Qwen2.5-3B</strong> </td> <td><strong>Qwen2.5-3B-quantized.w4a16 (this model)</strong> </td> <td><strong>Recovery</strong> </td> </tr> <tr> <td>MMLU (5-shot) </td> <td>65.68 </td> <td>64.10 </td> <td>97.6% </td> </tr> <tr> <td>ARC Challenge (25-shot) </td> <td>53.58 </td> <td>51.19 </td> <td>95.6% </td> </tr> <tr> <td>GSM-8K (5-shot, strict-match) </td> <td>68.23 </td> <td>67.85 </td> <td>99.4% </td> </tr> <tr> <td>Hellaswag (10-shot) </td> <td>74.46 </td> <td>73.36 </td> <td>98.5% </td> </tr> <tr> <td>Winogrande (5-shot) </td> <td>70.64 </td> <td>70.32 </td> <td>99.6% </td> </tr> <tr> <td>TruthfulQA (0-shot, mc2) </td> <td>48.93 </td> <td>46.22 </td> <td>94.2% </td> </tr> <tr> <td><strong>Average</strong> </td> <td><strong>63.59</strong> </td> <td><strong>62.18</strong> </td> <td><strong>97.8%</strong> </td> </tr> </table>