ai/deepseek-r1-distill-llama Docker 镜像 - 轩辕镜像

Deepseek-R1-Distill-Llama

!logo

DeepSeek introduced its first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1, leveraging reinforcement learning to enhance reasoning performance, with DeepSeek-R1 achieving state-of-the-art results and open-sourcing multiple distilled models.

The models provided here are the distill-llama variants, which are llama based models that have been fine-tuned on the responses and reasoning output of the full DeepSeek-R1 model.

Intended uses

Deepseek-R1-Distill-Llama can help with:

• Software development: Generates code, debugs, and explains complex concepts.
• Mathematics: Solves and explains complex problems for research and education.
• Content creation and editing: Writes, edits, and summarizes content for various industries.
• Customer service: Powers chatbots to engage users and answer queries.
• Data analysis: Extracts insights and generates reports from large datasets.
• Education: Acts as a digital tutor, providing clear explanations and personalized lessons.

Characteristics

i: Estimated

Available model variants

¹: VRAM estimated based on model characteristics.

latest → 8B-Q4_K_M

Use this AI model with Docker Model Runner

First, pull the model:

bash
docker model pull ai/deepseek-r1-distill-llama

Then run the model:

bash
docker model run ai/deepseek-r1-distill-llama

For more information on Docker Model Runner, explore the documentation.

Usage tips

• Set the temperature between 0.5 and 0.7 (recommended: 0.6) to avoid repetition or incoherence.
• Do not use a system prompt. Include all instructions within the user prompt.
• For math problems, add a directive like: "Please reason step by step and enclose the final answer in \boxed{}."

This model is sensitive to prompts. Few-shot prompting consistently degrades its performance. Therefore, we recommend you directly describe the problem and specify the output format using a zero-shot setting for optimal results.

Benchmark performance

Links

• DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning