DeepSeek R1 Distill Qwen 32B

DeepSeek R1 Distill Qwen 32B – Dense Reasoning Model via Knowledge Distillation

DeepSeek R1 Distill Qwen 32B is a 32B dense model produced by fine-tuning Qwen 2.5 32B on 800k reasoning samples curated from DeepSeek R1. It is one of six distilled models released alongside DeepSeek R1, and is the largest of the four Qwen-based distillations. It is officially documented as outperforming OpenAI o1-mini across various benchmarks, achieving new state-of-the-art results for dense models at its size.

Base Model & Licensing

Derived from Qwen 2.5 32B, originally licensed under Apache 2.0, and now fine-tuned with DeepSeek R1 reasoning data. Released under the MIT License, which explicitly permits commercial use, modifications, derivative works, and further distillation for training other LLMs.

Usage Recommendations

Officially documented usage configurations for the R1 Distill series:

• Temperature: 0.5–0.7 (0.6 recommended) to prevent endless repetition or incoherent outputs
• System prompt: Avoid adding a system prompt; all instructions should be in the user prompt
• Math tasks: Include a directive such as "Please reason step by step, and put your final answer within \boxed{}"
• Evaluation: Conduct multiple tests and average results.