Nour-Fayed/DeepSeek-R1-Distill-Qwen-32B

TEXT GENERATIONConcurrency Cost:2Model Size:32.8BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jul 2, 2026License:mitArchitecture:Transformer Open Weights Cold

The Nour-Fayed/DeepSeek-R1-Distill-Qwen-32B is a 32.8 billion parameter language model distilled from DeepSeek-R1, developed by DeepSeek-AI, with a 32768 token context length. It is fine-tuned based on Qwen2.5-32B using reasoning data generated by the larger DeepSeek-R1 model. This model excels in reasoning tasks across math, code, and general English and Chinese benchmarks, outperforming OpenAI-o1-mini in various evaluations.

Loading preview...

DeepSeek-R1-Distill-Qwen-32B Overview

This model is a 32.8 billion parameter distilled version of DeepSeek-R1, developed by DeepSeek-AI, built upon the Qwen2.5-32B base model. It leverages reasoning patterns from the larger DeepSeek-R1 model, which was developed using a novel large-scale reinforcement learning (RL) approach without initial supervised fine-tuning (SFT) to foster complex reasoning behaviors like self-verification and reflection. The distillation process transfers these advanced reasoning capabilities to a smaller, more efficient model.

Key Capabilities

  • Enhanced Reasoning: Demonstrates strong performance in complex reasoning tasks across mathematics, coding, and general knowledge.
  • Distilled Performance: Achieves competitive results against larger models like OpenAI-o1-mini, showcasing that smaller models can be powerful when distilled from advanced reasoning models.
  • Broad Benchmark Coverage: Evaluated across a wide range of benchmarks including AIME 2024, MATH-500, GPQA Diamond, LiveCodeBench, and CodeForces, with notable scores in math and code.
  • Context Length: Supports a context length of 32768 tokens.

Good For

  • Reasoning-intensive applications: Ideal for tasks requiring logical deduction, problem-solving, and multi-step thinking.
  • Resource-constrained environments: Offers strong performance in a 32.8B parameter footprint, making it suitable for scenarios where larger models are impractical.
  • Research and development: Provides a powerful base for further fine-tuning or experimentation with reasoning-focused LLMs.

Usage Recommendations

To achieve optimal performance, DeepSeek-AI recommends setting the temperature between 0.5-0.7 (0.6 recommended), avoiding system prompts, and including directives like "Please reason step by step, and put your final answer within \boxed{}" for mathematical problems. It is also advised to enforce the model to start its response with "\n" to ensure thorough reasoning.

Finetunes
1 models