Starred repositories
DeepSeek 4 Flash and PRO local inference engine for Metal, CUDA and ROCm
AOTInductor Weight Streaming from PyTorch Models
Full-parameter reward-model training sharded with FSDP2, verified on CPU (collective counts, bitwise resume from a sharded checkpoint), with a GPU sweep on 4x L40S planned for throughput, memory, s…
Personal daily InShorts-style digest for LLM/SLM inference-engineering news
Nauti-Strata: Speculative lookahead MoE scheduling, dynamic PCIe prefetching, and legacy compute (sm_60/AVX) acceleration
Jev-like family of decision models built on top of Qwen3.5/3.8 you can train and run on your own
Open-source Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO) platform. Track how ChatGPT, Claude, Perplexity, Gemini, Copilot, Grok, and Google AI Overviews mention and cit…
High-performance, SIMD-accelerated vector search engine and MCP server for .NET 10. Optimized for LLM embeddings and semantic search.
Fork of LM Evaluation Harness Suite for evaluating benchmarks in paper titled "PLDR-LLMs Learn A Generalizable Tensor Operator That Can Replace Its Own Deep Neural Net At Inference"
O(N·K) multi-head attention for PyTorch — a sparse drop-in replacement for dense scaled-dot-product attention.
The open-source CapCut alternative
GaetanLepage / tensordict
Forked from pytorch/tensordictTensorDict is a pytorch dedicated tensor container.
kakarot-dev / deepmiro
Forked from 666ghj/MiroFishOpen-source swarm intelligence engine — simulate hundreds of AI agents to predict outcomes. Includes MCP server for Claude.
# TurboQuant v3 (INT4 + AWQ + Protected Channels + Low-Rank) This notebook demonstrates a **TurboQuant-like** quantization algorithm: - Group-wise INT4 quantization - Activation-aware scaling (AWQ-…
Bash is all you need - A nano claude code–like 「agent harness」, built from 0 to 1
An adaptive continually learning hybrid RAG Agent for content, curation, and persona growth. It combines spaCy-based NLP, a persona graph, BM25 retrieval, a truth gate, confidence scoring, and loca…
An experimental pi extension that runs and manages qwen with llama.cpp
本地优先的跨平台 Claude Code / AI Agent 桌面工作台:多 Agent 协作、Git Worktree、代码 Diff、技能市场、多模型、Computer Use、桌面宠物;支持微信、飞书、钉钉及 H5 等远程访问。 | Local-first Claude Code / AI Agent desktop workspace: multi-agent teams, Git…
TokenSpeed is a speed-of-light LLM inference engine.
A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode,DeepSeek Harness and more. Only official website: ccswitch.io
Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Implementation of Inverse Reinforcement Learning (IRL) algorithms in Python/Tensorflow. Deep MaxEnt, MaxEnt, LPIRL
Autoresearch for GPU kernels. Give it any PyTorch model, go to sleep, wake up to optimized Triton kernels.
Maximum Entropy and Maximum Causal Entropy Inverse Reinforcement Learning Implementation in Python
Harness Engineering 学习指南 — 从概念理解到独立实践的深度学习档案