Tag

DeepGEMM

High-performance CUDA tensor-core GEMM kernel library for LLM workloads: supports FP8/FP4/BF16, fused Mega MoE and MQA scoring, and runtime JIT-compiled kernels. Targets NVIDIA SM90/SM100 and PyTorch—for teams working on low-level GPU kernel optimization.

DeepEP

Provides high-throughput, low-latency GPU communication kernels for Mixture-of-Experts (MoE) and expert-parallel workloads, with NVLink↔RDMA-aware forwarding, FP8/BF16 support, and low-latency RDMA hooks for inference decoding.

FlashMLA

Optimized MLA (Multi-head Latent Attention) decoding kernels powering DeepSeek-V3/V3.2 inference on Hopper and Blackwell GPUs. Dense decoding reaches ~3000 GB/s and 660 TFLOPS on H800; the sparse path stores the KV cache in FP8.

DeerFlow

Orchestrates a lead agent, isolated parallel sub-agents, long-term memory, and sandboxes for long-horizon tasks — minutes to hours of deep research, coding, and content creation. LangChain/LangGraph-based with extensible skills; v2 is a full rewrite.

VideoAgent

An agentic framework that analyzes, plans, and executes multi-step video understanding and editing workflows using multimodal LLM-driven agents—features intent decomposition, graph-based workflow orchestration, and automated shot planning for long-form video tasks.

DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models

An open large language model pairing DeepSeek Sparse Attention (DSA) for cheaper long-context inference with a scaled RL pipeline. Authors claim parity with GPT-5, with a high-compute Speciale variant surpassing it and rivaling Gemini-3.0-Pro on reasoning.

Engram — Conditional Memory via Scalable Lookup

Provides a conditional memory module that performs O(1) N‑gram lookups and fuses static embeddings into transformer hidden states — enables offloading large embedding tables to host memory with minimal inference overhead.

CodeWhale

Terminal-native coding agent that streams reasoning blocks, makes controlled edits to local workspaces behind approval gates, and includes an auto mode that chooses model and thinking level per turn — designed for in-terminal code review, debugging, and automation workflows.

DeepSeek TUI

Terminal-native coding agent that runs DeepSeek V4 models with a 1M-token context; provides file, shell, git and web tools, thinking-mode streaming, a sandbox and durable task queue — single binary with no Node/Python runtime.

DS2API

Converts DeepSeek protocol calls into OpenAI/Claude/Gemini-compatible APIs with a Go backend and React admin UI. Offers account pooling, protocol adapters, tool-call translation, PoW, and multiple deployment modes (Docker, Vercel, standalone).

Nemotron-SFT-Math-v3

Large-scale mathematical reasoning dataset of model-generated solution trajectories produced with and without Python Tool-Integrated Reasoning (TIR), with final answers verified against reference solutions. Contains ~3.64M JSONL training samples (~144 GB) and per-source CC-BY / CC-BY-SA licensing; intended for training and evaluating tool-augmented mathematical reasoning in LLMs.

BugHunter

Terminal-first toolkit that automates bug bounty workflows — recon, hunting across 20 vulnerability classes, validation, and submission-ready report generation; runs as a Claude Code plugin or standalone CLI with support for free local AI providers (Ollama, Groq, DeepSeek).