Qwen2.5-32B-quantized.w8a16

Qwen2.5-32B-quantized.w8a16

Model Overview

• Model Architecture: Qwen2
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: INT8
• Intended Use Cases: Similarly to Qwen2.5-32B, this is a base language model.
• Out-of-scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws).
• Release Date: 10/09/2024
• Version: 1.0
• Model Developers: Neural Magic

Quantized version of Qwen2.5-32B. It achieves an OpenLLMv1 score of 75.4, compared to 75.3 for Qwen2.5-32B.

Model Optimizations

This model was obtained by quantizing the weights of Qwen2.5-32B to INT8 data type. 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 of the linear operators within transformers blocks are quantized. Symmetric per-channel quantization is applied, in which a linear scaling per output dimension maps the INT8 and floating point representations of the quantized weights. The GPTQ algorithm is applied for quantization, as implemented in the llm-compressor library.

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/Qwen2.5-32B-quantized.w8a16"
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 OpenLLMv1 benchmark, composed of MMLU, ARC-Challenge, GSM-8K, Hellaswag, Winogrande and TruthfulQA. Evaluation was conducted using lm-evaluation-harness and the vLLM engine.

Accuracy

<table> <tr> <td><strong>Category</strong> </td> <td><strong>Benchmark</strong> </td> <td><strong>Qwen2.5-32B</strong> </td> <td><strong>Qwen2.5-32B-quantized.w8a16<br>(this model)</strong> </td> <td><strong>Recovery</strong> </td> </tr> <tr> <td rowspan="8" ><strong>OpenLLM v1</strong> </td> </tr> <tr> <td>MMLU (5-shot) </td> <td>83.25 </td> <td>83.19 </td> <td>99.9% </td> </tr> <tr> <td>ARC Challenge (25-shot) </td> <td>66.30 </td> <td>66.04 </td> <td>99.6% </td> </tr> <tr> <td>GSM-8k (5-shot, strict-match) </td> <td>78.09 </td> <td>78.62 </td> <td>100.7% </td> </tr> <tr> <td>Hellaswag (10-shot) </td> <td>85.08 </td> <td>85.14 </td> <td>100.1% </td> </tr> <tr> <td>Winogrande (5-shot) </td> <td>81.29 </td> <td>81.61 </td> <td>100.4% </td> </tr> <tr> <td>TruthfulQA (0-shot, mc2) </td> <td>57.76 </td> <td>57.78 </td> <td>100.0% </td> </tr> <tr> <td><strong>Average</strong> </td> <td><strong>75.30</strong> </td> <td><strong>75.40</strong> </td> <td><strong>100.1%</strong> </td> </tr> </table>

Reproduction

The results were obtained using the following command:

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Qwen2.5-32B-quantized.w8a16",dtype=auto,max_model_len=4096,add_bos_token=True,tensor_parallel_size=1 \
  --tasks openllm \
  --batch_size auto