Zhihong Shao

Here comes DeepSeek-R1, our latest reasoning model with significantly enhanced reasoning abilities. We also share a technical report on how we train the reasoning model with large-scale RL. Have fun!

We just released DeepSeek-Prover V2. - Solves nearly 90% of miniF2F problems - Significantly improves the SoTA performance on the PutnamBench - Achieves a non-trivial pass rate on AIME 24 & 25 problems in their formal version Github: github.com/deepseek-ai/De…

🚀 Day 6 of #OpenSourceWeek: One More Thing – DeepSeek-V3/R1 Inference System Overview Optimized throughput and latency via: 🔧 Cross-node EP-powered batch scaling 🔄 Computation-communication overlap ⚖️ Load balancing Statistics of DeepSeek's Online Service: ⚡ 73.7k/14.8k…

🚀 Day 5 of #OpenSourceWeek: 3FS, Thruster for All DeepSeek Data Access Fire-Flyer File System (3FS) - a parallel file system that utilizes the full bandwidth of modern SSDs and RDMA networks. ⚡ 6.6 TiB/s aggregate read throughput in a 180-node cluster ⚡ 3.66 TiB/min…

🚀 Day 4 of #OpenSourceWeek: Optimized Parallelism Strategies ✅ DualPipe - a bidirectional pipeline parallelism algorithm for computation-communication overlap in V3/R1 training. 🔗 github.com/deepseek-ai/Du… ✅ EPLB - an expert-parallel load balancer for V3/R1. 🔗…

🚀 Day 3 of #OpenSourceWeek: DeepGEMM Introducing DeepGEMM - an FP8 GEMM library that supports both dense and MoE GEMMs, powering V3/R1 training and inference. ⚡ Up to 1350+ FP8 TFLOPS on Hopper GPUs ✅ No heavy dependency, as clean as a tutorial ✅ Fully Just-In-Time compiled…

🚀 Day 2 of #OpenSourceWeek: DeepEP Excited to introduce DeepEP - the first open-source EP communication library for MoE model training and inference. ✅ Efficient and optimized all-to-all communication ✅ Both intranode and internode support with NVLink and RDMA ✅…

🚀 Day 1 of #OpenSourceWeek: FlashMLA Honored to share FlashMLA - our efficient MLA decoding kernel for Hopper GPUs, optimized for variable-length sequences and now in production. ✅ BF16 support ✅ Paged KV cache (block size 64) ⚡ 3000 GB/s memory-bound & 580 TFLOPS…

🚀 Day 0: Warming up for #OpenSourceWeek! We're a tiny team @deepseek_ai exploring AGI. Starting next week, we'll be open-sourcing 5 repos, sharing our small but sincere progress with full transparency. These humble building blocks in our online service have been documented,…

Congrats to DeepSeek on producing an o1-level reasoning model! Their research paper demonstrates that they’ve independently found some of the core ideas that we did on our way to o1.

Again guys, this is not our Wenfeng.

DeepSeek-R1 (Preview) Results 🔥 We worked with the @deepseek_ai team to evaluate R1 Preview models on LiveCodeBench. The model performs in the vicinity of o1-Medium providing SOTA reasoning performance! Huge kudos to the team and I'm looking forward to the full release!! /1

Here comes the official release of DeepSeek-V3. We also share a lot in the tech report. Check it out!

Our DeepSeek reasoning model is great on code and math. Try it out!

o1 — our first model trained with reinforcement learning to think hard about problems before answering. Extremely proud of the team! This is a new paradigm with vast opportunity. This is evident quantitatively (eg reasoning metrics are already a step function improved) and…

I have always believed that you don't need a GPT-6 quality base model to achieve human-level reasoning performance, and that reinforcement learning was the missing ingredient on the path to AGI. Today, we have the proof -- o1. x.com/OpenAI/status/…

Super interesting work from DeepSeek on MiniF2F (so happy to see our benchmark still in use \o/). It's hard to compare this with the recent DeepMind paper but from my experience building and using MiniF2F I think ~60% pass-rate is likely comparable to DeepMind's recent result on…

DeepSeek-Prover-V1.5: Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search > New SOTA: 63.5% on miniF2F (high school) & 25.3% on ProofNet (undergrad) > Introduces RMaxTS: Novel MCTS for diverse proof generation > Features RLPAF:…

LLMs can assist humans in providing feedback to train the next LLM. Our recent work led by @jiaxinwen22 shows that LLMs can empower non-experts to match expert programmers in fixing LLM-generated code on the competitive programming task.

📢 After 3 months, the AI Mathematical Olympiad (AIMO) on Kaggle has announced the winners! 🎉 We're thrilled to see the Top 4 teams all chose DeepSeekMath-7B as their base model, with Numina @JiaLi52524397 achieving 29/50 correct answers! 👏 Even Terence Tao was amazed. 🤯…