---

Hello, DeepSeek Open Infra!

202502 Open-Source Week

We're a tiny team @deepseek-ai pushing our limits in AGI exploration.

Starting this week , Feb 24, 2025 we'll open-source 5 repos – one daily drop – not because we've made grand claims, but simply as developers sharing our small-but-sincere progress with full transparency.

These are humble building blocks of our online service: documented, deployed, and battle-tested in production. No vaporware, just sincere code that moved our tiny yet ambitious dream forward.

Why? Because every line shared becomes collective momentum that accelerates the journey. Daily unlocks begin soon. No ivory towers - just pure garage-energy and community-driven innovation 🔧

Stay tuned – let's geek out in the open together.

Day 1 - FlashMLA

Efficient MLA Decoding Kernel for Hopper GPUs
Optimized for variable-length sequences, battle-tested in production

🔗 FlashMLA GitHub Repo
✅ BF16 support
✅ Paged KV cache (block size 64)
⚡ Performance: 3000 GB/s memory-bound | BF16 580 TFLOPS compute-bound on H800

Day 2 - DeepEP

Excited to introduce DeepEP - the first open-source EP communication library for MoE model training and inference.

🔗 DeepEP GitHub Repo
✅ Efficient and optimized all-to-all communication
✅ Both intranode and internode support with NVLink and RDMA
✅ High-throughput kernels for training and inference prefilling
✅ Low-latency kernels for inference decoding
✅ Native FP8 dispatch support
✅ Flexible GPU resource control for computation-communication overlapping

Day 3 - DeepGEMM

Introducing DeepGEMM - an FP8 GEMM library that supports both dense and MoE GEMMs, powering V3/R1 training and inference.

🔗 DeepGEMM GitHub Repo
⚡ Up to 1350+ FP8 TFLOPS on Hopper GPUs
✅ No heavy dependency, as clean as a tutorial
✅ Fully Just-In-Time compiled
✅ Core logic at ~300 lines - yet outperforms expert-tuned kernels across most matrix sizes
✅ Supports dense layout and two MoE layouts

Day 4 - Optimized Parallelism Strategies

✅ DualPipe - a bidirectional pipeline parallelism algorithm for computation-communication overlap in V3/R1 training.
🔗 GitHub Repo

✅ EPLB - an expert-parallel load balancer for V3/R1.
🔗 GitHub Repo

📊 Analyze computation-communication overlap in V3/R1.
🔗 GitHub Repo

Day 5 - 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 throughput on GraySort benchmark in a 25-node cluster
⚡ 40+ GiB/s peak throughput per client node for KVCache lookup
🧬 Disaggregated architecture with strong consistency semantics
✅ Training data preprocessing, dataset loading, checkpoint saving/reloading, embedding vector search & KVCache lookups for inference in V3/R1

📥 3FS → 🔗GitHub Repo
⛲ Smallpond - data processing framework on 3FS → 🔗GitHub Repo

Day 6 - 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

Production data of V3/R1 online services:
⚡ 73.7k/14.8k input/output tokens per second per H800 node
🚀 Cost profit margin 545%

💡 We hope this week's insights offer value to the community and contribute to our shared AGI goals.

📖 Deep Dive: 🔗Day 6 - One More Thing: DeepSeek-V3/R1 Inference System Overview
📖 中文版: 🔗DeepSeek-V3 / R1 推理系统概览

2024 AI Infrastructure Paper (SC24)

Fire-Flyer AI-HPC: A Cost-Effective Software-Hardware Co-Design for Deep Learning

📄 Paper Link
📄 Arxiv Paper Link