arxiv Moonshot AI (arXiv) · 635d ago

DeepSeek-V3: 671B MoE model with auxiliary-loss-free balancing

from English

DeepSeek-AI presents DeepSeek-V3, a strong Mixture-of-Experts language model with 671B total parameters and 37B activated per token. The model adopts Multi-head Latent Attention and DeepSeekMoE architectures to achieve efficient inference and cost-effective training.

DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing to minimize performance degradation.
It employs a multi-token prediction training objective to enhance overall benchmark performance.
The team validates FP8 mixed precision training on an extremely large-scale model for the first time.
Training utilized 14.8 trillion tokens and cost only 2.788M H800 GPU hours, totaling approximately $5.576 million.
Evaluations show it outperforms other open-source models and matches leading closed-source models like GPT-4o and Claude-3.5-Sonnet.

The report emphasizes that DeepSeek-V3 achieves strong performance comparable to top-tier closed-source models while maintaining remarkably low training costs and stability.

Importance 3/3 Moonshot AI (arXiv) DeepSeek Research paper Training methods

Read original

arxiv Moonshot AI (arXiv) · 609d ago · 16 views

DeepSeek-R1 incentivizes LLM reasoning via reinforcement learning

DeepSeek has released DeepSeek-R1, a large language model trained using pure reinforcement learning to enhance its reasoning capabilities without relying on human-annotated demonstrations.

arxiv Moonshot AI (arXiv) · 609d ago · 32 views

DeepSeek releases R1 reasoning models via reinforcement learning

DeepSeek introduces its first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1, which achieve performance comparable to OpenAI-o1-1217 on reasoning tasks. DeepSeek-R1-Zero is trained via large-scale reinforcement learning without supervised fine-tuning, while DeepSeek-R1 uses a multi-stage pipeline with cold-start data to improve readability.

arxiv arXiv cs.LG · 1d ago · 10 views

Fathom-Vaidya improves medical reasoning with rubric-based rewards

The authors introduce Fathom-Vaidya, a 30B parameter model that uses synthetic data and rubric-based reinforcement learning to enhance diagnostic and clinical healthcare reasoning. The training framework first applies rule-guided RL to MedBullets-derived questions for diagnosis, then utilizes 5.3k synthetic multi-turn scenarios with multi-dimensional rubrics for interactive clinical tasks.

arxiv arXiv cs.AI · 1d ago · 12 views

Fathom-Vaidya improves medical reasoning with rubric-based rewards

The authors introduce Fathom-Vaidya, a 30B parameter model that uses synthetic data and rubric-based reinforcement learning to enhance both diagnostic and clinical healthcare reasoning. The training framework first targets diagnostic accuracy using MedBullets-derived questions and then addresses multi-turn clinical interactions through 5.3k generated scenarios with multi-dimensional rubrics.

arxiv arXiv cs.CL · 1d ago · 12 views

Fathom-Vaidya improves medical reasoning with rubric-based rewards

The authors introduce Fathom-Vaidya, a 30B parameter model that uses synthetic data and rubric-based reinforcement learning to enhance both diagnostic and clinical healthcare reasoning.