deepseek-r1-distill-qwen-32b-gptq-int4

DeepSeek-R1-Distill-Qwen-32B-GPTQ-Int4

原始模型 deepseek-ai/DeepSeek-R1-Distill-Qwen-32B

💡💡【旧显卡运行，回答全是感叹号，需要做些什么吗？】💡💡

这个不是模型问题，gptq的cuda算子有针对int4和int8分别在内容少于 50 和 24 tokens的时候，走特殊的处理和计算方式 源码

这个优化方式，没有处理好浮点稳定性的问题，导致感叹号输出。

解决方法就是:

• 换新显卡，RTX 30系列 / Ampere架构同期及之后的显卡，自动支持 gptq marlin 运算，则没有这个问题。或
• 停用这个特殊优化，把 50 和 24 都设置成 0 ，但这需要重新编译gptq对应cuda代码。或
• 🔥🔥🔥绕过这个特殊优化🔥🔥🔥，在 tokenizer config.json 的 chat template 处，在字符串左端，注入垃圾字符。这样就是在内化的模板层面，强行顶上 token 数，就不会走这个特殊优化路径了。

例如：

【模型更新日期】

注：通过 snapshot download 函数传入 revision=... 来下载指定的 tag 版本

【模型列表】

tag 文件大小 最近更新时间

• -------- -------- --------------

g128 19GB 2025-01-21

【模型下载】

【修复内容】

• 对GPTQ量化的校准做了额外优化；减少模型的 1.乱吐字 、 2.无限循环 、 3.长文能力丢失 等情况。
• 根据模型实际情况，可以支持1卡、2卡及4卡的 tensor-parallel-size 启动。

【介绍】

• Introduction

We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1.

DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrated remarkable performance on reasoning.

With RL, DeepSeek-R1-Zero naturally emerged with numerous powerful and interesting reasoning behaviors.

However, DeepSeek-R1-Zero encounters challenges such as endless repetition, poor readability, and language mixing. To address these issues and further enhance reasoning performance,

we introduce DeepSeek-R1, which incorporates cold-start data before RL.

DeepSeek-R1 achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks.

To support the research community, we have open-sourced DeepSeek-R1-Zero, DeepSeek-R1, and six dense models distilled from DeepSeek-R1 based on Llama and Qwen. DeepSeek-R1-Distill-Qwen-32B outperforms OpenAI-o1-mini across various benchmarks, achieving new state-of-the-art results for dense models.

• Model Summary

Post-Training: Large-Scale Reinforcement Learning on the Base Model**

We directly apply reinforcement learning (RL) to the base model without relying on supervised fine-tuning (SFT) as a preliminary step. This approach allows the model to explore chain-of-thought (CoT) for solving complex problems, resulting in the development of DeepSeek-R1-Zero. DeepSeek-R1-Zero demonstrates capabilities such as self-verification, reflection, and generating long CoTs, marking a significant milestone for the research community. Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT. This breakthrough paves the way for future advancements in this area.

We introduce our pipeline to develop DeepSeek-R1. The pipeline incorporates two RL stages aimed at discovering improved reasoning patterns and aligning with human preferences, as well as two SFT stages that serve as the seed for the model's reasoning and non-reasoning capabilities.

We believe the pipeline will benefit the industry by creating better models.

Distillation: Smaller Models Can Be Powerful Too**

We demonstrate that the reasoning patterns of larger models can be distilled into smaller models, resulting in better performance compared to the reasoning patterns discovered through RL on small models. The open source DeepSeek-R1, as well as its API, will benefit the research community to distill better smaller models in the future.

Using the reasoning data generated by DeepSeek-R1, we fine-tuned several dense models that are widely used in the research community. The evaluation results demonstrate that the distilled smaller dense models perform exceptionally well on benchmarks. We open-source distilled 1.5B, 7B, 8B, 14B, 32B, and 70B checkpoints based on Qwen2.5 and Llama3 series to the community.

【高并发RESTFul API推理】

方式1：vllm

方式2: sglang