Nour-Fayed/DeepSeek-R1-Distill-Llama-70B

TEXT GENERATIONConcurrency Cost:4Model Size:70BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jul 2, 2026License:mitArchitecture:Transformer Open Weights Cold

Nour-Fayed/DeepSeek-R1-Distill-Llama-70B is a 70 billion parameter language model distilled from DeepSeek-R1, developed by DeepSeek-AI. This model is fine-tuned using reasoning data generated by the larger DeepSeek-R1, which itself was trained with large-scale reinforcement learning to enhance reasoning capabilities. It excels in complex reasoning tasks across math, code, and general English and Chinese benchmarks, demonstrating strong performance comparable to larger proprietary models. Optimized for robust reasoning, it is suitable for applications requiring advanced problem-solving and logical inference.

Loading preview...

DeepSeek-R1-Distill-Llama-70B: Reasoning-Enhanced LLM

This model is a 70 billion parameter variant from the DeepSeek-R1-Distill series, developed by DeepSeek-AI. It is a distilled version of the powerful DeepSeek-R1, which was uniquely trained using large-scale reinforcement learning (RL) without initial supervised fine-tuning (SFT) to foster advanced reasoning behaviors like self-verification and reflection. The distillation process transfers these sophisticated reasoning patterns from the larger DeepSeek-R1 into smaller, dense models, including this Llama-based 70B checkpoint.

Key Capabilities

  • Advanced Reasoning: Inherits and demonstrates strong reasoning capabilities across math, code, and general knowledge tasks, derived from the DeepSeek-R1 parent model.
  • Performance: Achieves competitive results on benchmarks such as AIME 2024 (70.0 pass@1), MATH-500 (94.5 pass@1), GPQA Diamond (65.2 pass@1), and LiveCodeBench (57.5 pass@1), often outperforming other models in its class.
  • Distillation Advantage: Leverages reasoning data generated by DeepSeek-R1 to enhance performance in a more compact architecture, proving that complex reasoning can be effectively distilled.
  • Extended Context: Supports a context length of 32,768 tokens.

Usage Recommendations

  • Prompting: Avoid system prompts; include all instructions within the user prompt. For mathematical problems, use directives like "Please reason step by step, and put your final answer within \boxed{}".
  • Temperature: Recommended temperature range is 0.5-0.7 (0.6 is ideal) to prevent repetitive or incoherent outputs.
  • Enforce Reasoning: To ensure thorough reasoning, it is recommended to enforce the model to initiate its response with "\n".

Finetunes
1 models