DeepSeek ORG

No description.

No description.

DeepSeek-V3 is an open-weights AI language model with 671 billion parameters whose performance rivals top closed-source models, released under an open license so you can run it locally or integrate it into your own product.

No description.

DeepSeek-R1 is a family of open AI reasoning models trained with reinforcement learning to think step-by-step through maths and code, matching OpenAI-o1 performance, with smaller distilled versions for less powerful hardware.

Integrate the DeepSeek API into popular software

A curated directory of third-party apps, plugins, and frameworks that integrate DeepSeek AI, helping you find existing tools instead of building from scratch.

DeepSeek Coder: Let the Code Write Itself

Family of AI models trained on code to complete, insert, and generate code across 80+ programming languages. Self-hosted alternative to cloud-based coding assistants.

Contexts Optical Compression

AI model that extracts text from images and documents using efficient vision processing, converting photos and PDFs into readable text or formatted Markdown.

Janus-Series: Unified Multimodal Understanding and Generation Models

A family of AI models from DeepSeek that can both describe what is in an image and generate new images from text prompts, all within a single unified model rather than needing two separate tools.

FlashMLA: Efficient Multi-head Latent Attention Kernels

A high-performance distributed file system designed to address the challenges of AI training and inference workloads.

DeepEP: an efficient expert-parallel communication library

Production-tested AI infrastructure tools for efficient AGI development and community-driven innovation

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

DeepSeek LLM: Let there be answers

DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence

DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding

DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model

A lightweight data processing framework built on DuckDB and 3FS.

Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models

DeepSeek-VL: Towards Real-World Vision-Language Understanding

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

[ICLR 2024] Official implementation of DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior

Visual Causal Flow

A bidirectional pipeline parallelism algorithm for computation-communication overlap in DeepSeek V3/R1 training.

DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models

No description.

A kernel library written in tilelang

No description.

Expert Parallelism Load Balancer

No description.

Analyze computation-communication overlap in V3/R1.

A curated list of open-source projects related to DeepSeek Coder

Expert Specialized Fine-Tuning

No description.

An early research stage expert-parallel load balancer for MoE models based on linear programming.