QuantTrio/Qwen3-235B-A22B-GPTQ-Int8

通义千问3-235B-A22B-GPTQ-Int8

基础型 Qwen/Qwen3-235B-A22B

【模型更新日期】

2025-05-09
1. 首次commit
2. 确定支持8卡的`tensor-parallel-size` + `expert-parallel` 启动
3. 必须 gptq_marlin 启动；不支持 Compute 7 显卡启动：vllm没有实现原生GPTQ的moe模块。

【依赖】

vllm==0.8.5
transformers==4.51.3

### 【💡新版 VLLM MoE 注意事项💡】

1. 需使用V0推理模式

启动vllm之前，先设置环境变量

export VLLM_USE_V1=0

2. gptq_marlin.py 存在小bug，需要补丁

将附件中的文件替换至

.../vllm/model_executor/layers/quantization/gptq_marlin.py

否则会出现下述报错

    raise NotImplementedError(
NotImplementedError: Apply router weight on input is not supported forfused Marlin MoE method.

### 【💡通义千问3-235B-A22B 注意事项💡】

1. 启动vllm的时候，要记得使用专家并行模式(--enable-expert-parallel)，否则不能单节点8卡启动。

启动示例：

vllm serve \
    tclf90/Qwen3-235B-A22B-GPTQ-Int8 \
    --served-model-name Qwen3-235B-A22B-GPTQ-Int8 \
    --max-num-seqs 8 \
    --max-model-len 32768 \
    --max-seq-len-to-capture 32768 \
    --gpu-memory-utilization 0.98 \
    --tensor-parallel-size 8 \
    --enable-expert-parallel \
    --disable-log-requests \
    --trust-remote-code

【模型列表】

文件大小	最近更新时间
226GB	2025-05-09

【模型下载】

from modelscope import snapshot_download
snapshot_download('tclf90/Qwen3-235B-A22B-GPTQ-Int8', cache_dir="本地路径")

【介绍】

Qwen3-235B-A22B

Qwen3 Highlights

Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support, with the following key features:

• Uniquely support of seamless switching between thinking mode (for complex logical reasoning, math, and coding) and non-thinking mode (for efficient, general-purpose dialogue) within single model, ensuring optimal performance across various scenarios.
• Significantly enhancement in its reasoning capabilities, surpassing previous QwQ (in thinking mode) and Qwen2.5 instruct models (in non-thinking mode) on mathematics, code generation, and commonsense logical reasoning.
• Superior human preference alignment, excelling in creative writing, role-playing, multi-turn dialogues, and instruction following, to deliver a more natural, engaging, and immersive conversational experience.
• Expertise in agent capabilities, enabling precise integration with external tools in both thinking and unthinking modes and achieving leading performance among open-source models in complex agent-based tasks.
• Support of 100+ languages and dialects with strong capabilities for multilingual instruction following and translation.

Model Overview

Qwen3-235B-A22B has the following features:

• Type: Causal Language Models
• Training Stage: Pretraining & Post-training
• Number of Parameters: 235B in total and 22B activated
• Number of Paramaters (Non-Embedding): 234B
• Number of Layers: 94
• Number of Attention Heads (GQA): 64 for Q and 4 for KV
• Number of Experts: 128
• Number of Activated Experts: 8
• Context Length: 32,768 natively and 131,072 tokens with YaRN.

For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our blog, GitHub, and Documentation.

Best Practices

To achieve optimal performance, we recommend the following settings:

1. Sampling Parameters:

   • For thinking mode (enable_thinking=True), use Temperature=0.6, TopP=0.95, TopK=20, and MinP=0. DO NOT use greedy decoding, as it can lead to performance degradation and endless repetitions.
   • For non-thinking mode (enable_thinking=False), we suggest using Temperature=0.7, TopP=0.8, TopK=20, and MinP=0.
   • For supported frameworks, you can adjust the presence_penalty parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.

2. Adequate Output Length: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 38,912 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.

3. Standardize Output Format: We recommend using prompts to standardize model outputs when benchmarking.

   • Math Problems: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
   • Multiple-Choice Questions: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the answer field with only the choice letter, e.g., "answer": "C"."

4. No Thinking Content in History: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. It is implemented in the provided chat template in Jinja2. However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.

Processing Long Texts

Qwen3 natively supports context lengths of up to 32,768 tokens. For conversations where the total length (including both input and output) significantly exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively. We have validated the model's performance on context lengths of up to 131,072 tokens using the YaRN method.

YaRN is currently supported by several inference frameworks, e.g., transformers and llama.cpp for local use, vllm and sglang for deployment. In general, there are two approaches to enabling YaRN for supported frameworks:

• Modifying the model files: In the config.json file, add the rope_scaling fields:

  {
      ...,
      "rope_scaling": {
          "rope_type": "yarn",
          "factor": 4.0,
          "original_max_position_embeddings": 32768
      }
  }
  

  For llama.cpp, you need to regenerate the GGUF file after the modification.

• Passing command line arguments:

  For vllm, you can use

  vllm serve ... --rope-scaling '{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}' --max-model-len 131072  
  

  For sglang, you can use

  python -m sglang.launch_server ... --json-model-override-args '{"rope_scaling":{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}}'
  

  For llama-server from llama.cpp, you can use

  llama-server ... --rope-scaling yarn --rope-scale 4 --yarn-orig-ctx 32768
  

To completely disable thinking, you could use a custom chat template when starting the model:

Vllm Guide

vllm serve ...model_path... --chat-template ./qwen3_nonthinking.jinja

Citation

If you find our work helpful, feel free to give us a cite.

@misc{qwen3,
    title  = {Qwen3},
    url    = {https://qwenlm.github.io/blog/qwen3/},
    author = {Qwen Team},
    month  = {April},
    year   = {2025}
}