DeepSeek-R1 · GitHub Models · GitHub

DeepSeek-R1 stands out at thinking jobs using a step-by-step training procedure, such as language, scientific reasoning, and coding tasks. It features 671B overall specifications with 37B active parameters, and 128k context length.

DeepSeek-R1 builds on the development of earlier reasoning-focused designs that enhanced efficiency by extending Chain-of-Thought (CoT) reasoning. DeepSeek-R1 takes things further by combining reinforcement knowing (RL) with fine-tuning on carefully picked datasets. It evolved from an earlier version, DeepSeek-R1-Zero, which relied exclusively on RL and showed strong thinking skills but had problems like hard-to-read outputs and language disparities. To resolve these restrictions, DeepSeek-R1 incorporates a little amount of cold-start information and follows a refined training pipeline that blends reasoning-oriented RL with monitored fine-tuning on curated datasets, resulting in a model that achieves modern performance on reasoning standards.

Usage Recommendations

We to the following configurations when using the DeepSeek-R1 series models, including benchmarking, to achieve the expected performance:

– Avoid including a system timely; all instructions should be consisted of within the user prompt.
– For mathematical problems, it is suggested to consist of a regulation in your timely such as: “Please reason action by step, and put your final answer within boxed .”.
– When evaluating model efficiency, it is suggested to carry out several tests and average the results.

Additional recommendations

The design’s reasoning output (consisted of within the tags) may consist of more damaging material than the design’s last response. Consider how your application will utilize or show the thinking output; you might wish to reduce the reasoning output in a production setting.