DeepSeek Open-Sources DeepSeek-R1 LLM with Performance Comparable To OpenAI's O1 Model

DeepSeek open-sourced DeepSeek-R1, an LLM fine-tuned with support learning (RL) to improve thinking ability. DeepSeek-R1 attains results on par with OpenAI's o1 model on several criteria, including MATH-500 and SWE-bench.

DeepSeek-R1 is based on DeepSeek-V3, 89u89.com a mixture of specialists (MoE) model just recently open-sourced by DeepSeek. This base design is fine-tuned using Group Relative Policy Optimization (GRPO), a reasoning-oriented variation of RL. The research study group also performed knowledge distillation from DeepSeek-R1 to open-source Qwen and Llama designs and released a number of versions of each; these models surpass bigger designs, consisting of GPT-4, on math and coding criteria.

[DeepSeek-R1 is] the very first action towards enhancing language design reasoning abilities utilizing pure reinforcement learning (RL). Our objective is to check out the potential of LLMs to develop thinking capabilities without any supervised data, focusing on their self-evolution through a pure RL process...DeepSeek-R1 ... master a large range of tasks, including creative writing, basic concern answering, modifying, summarization, and more. Additionally, DeepSeek-R1 shows impressive efficiency on jobs needing long-context understanding, significantly outperforming DeepSeek-V3 on long-context criteria.

To establish the model, DeepSeek began with DeepSeek-V3 as a base. They first attempted fine-tuning it only with RL, and without any supervised fine-tuning (SFT), producing a model called DeepSeek-R1-Zero, which they have likewise launched. This design exhibits strong thinking performance, but" powerful thinking behaviors, it faces a number of problems. For instance, DeepSeek-R1-Zero struggles with challenges like bad readability and language mixing."

To address this, the team used a brief phase of SFT to prevent the "cold start" issue of RL. They collected several thousand examples of chain-of-thought thinking to utilize in SFT of DeepSeek-V3 before running RL. After the RL process converged, they then collected more SFT information utilizing rejection sampling, it-viking.ch resulting in a dataset of 800k samples. This dataset was utilized for further fine-tuning and to produce the distilled designs from Llama and Qwen.

DeepSeek examined their design on a variety of reasoning, math, and coding criteria and compared it to other designs, including Claude-3.5- Sonnet, GPT-4o, and o1. DeepSeek-R1 outperformed all of them on numerous of the benchmarks, consisting of AIME 2024 and hb9lc.org MATH-500.

DeepSeek-R1 Performance. Image Source: DeepSeek-R1 Technical Report

Within a couple of days of its release, the LMArena announced that DeepSeek-R1 was ranked # 3 general in the arena and raovatonline.org # 1 in coding and mathematics. It was likewise tied for # 1 with o1 in "Hard Prompt with Style Control" category.

Django structure co-creator Simon Willison wrote about his experiments with among the DeepSeek distilled Llama designs on his blog:

Each action begins with a ... pseudo-XML tag containing the chain of thought utilized to help produce the reaction. [Given the timely] "a joke about a pelican and a walrus who run a tea space together" ... It then believed for 20 paragraphs before outputting the joke! ... [T] he joke is horrible. But the procedure of arriving was such a fascinating insight into how these new designs work.

Andrew Ng's newsletter The Batch blogged about DeepSeek-R1:

DeepSeek is rapidly becoming a strong builder of open designs. Not only are these designs fantastic entertainers, but their license allows use of their outputs for distillation, potentially pushing forward the cutting-edge for language models (and multimodal designs) of all sizes.

The DeepSeek-R1 designs are available on HuggingFace.

About the Author