Embodied AI2025
Provides a unified Python interface to collect data, train visual/dynamics world models, and evaluate them with model-predictive control across many standardized environments. Includes reference baselines, planning solvers, dataset converters, and LanceDB-backed formats for reproducible experiments. Best suited for researchers benchmarking world-model algorithms.
AI Infra2025
Stores a pruned proximity graph instead of all embeddings, recomputing vectors on demand at query time. A 60M-doc index takes 6GB, not 201GB (97% less), at comparable recall. Powers private local RAG over files, mail, chat, and browser history.
AI Train2025
Forecasts financial candlesticks (OHLCV K-lines) with a decoder-only transformer pre-trained on 12B+ records from 45 exchanges. A tokenizer turns market data into discrete tokens, enabling price/volatility forecasting and synthetic K-line generation.
AI Infra2025
Model-compression toolkit for large LLMs/VLMs that integrates quantization (FP8/INT4/etc.), speculative decoding, token pruning and deployment hooks—designed for end-to-end performance on single/multi-GPU inference workflows and research-to-prod model optimization.
MCP Server2025
Provides a long‑lived, in‑process file and content search library for editors and AI agents, with typo‑resistant fuzzy matching, frecency‑ranked results, background watchers, and a lightweight in‑memory content index — optimized for repeated searches in long‑running processes.
AI Deploy2025
A ~5,000-line Python LLM inference engine that re-implements SGLang's serving optimizations — radix KV-cache reuse, chunked prefill, overlap scheduling, tensor parallelism — as a fully type-annotated reference instead of a black box.
AI Infra2025
Extends vLLM beyond text to serve omni-modal models — Qwen3-Omni, TTS like CosyVoice3, and diffusion image/video/audio generators — in one engine, adding the non-autoregressive Diffusion Transformer support the core project never targeted.
AI Others2025
Automatically removes safety alignment from transformer LLMs via directional ablation, with Optuna's TPE optimizer tuning the parameters — no retraining or model-internals expertise needed; hit 3/100 refusals at 0.16 KL on Gemma-3-12b.
AI Agent2025
Lets AI agents describe interactive UIs as declarative JSON instead of executable code; client apps render the components with native widgets from a pre-approved catalog, keeping agent-generated UI safe across trust boundaries.
AI Infra2025
Provides a Gymnasium-style API and tooling to create, deploy, and interact with isolated execution environments for agentic RL training. Includes async/sync clients, a web interface, CLI, Docker-based deployment, and Hugging Face Spaces integration.
AI Train2025
Worked examples and reusable abstractions for fine-tuning open LLMs via the Tinker training API: you write the training loop while distributed execution runs remotely. Covers SFT, math/code RL, DPO, three-stage RLHF, distillation, and tool use.
AI Coding2025
Turns documentation sites, GitHub repos, PDFs, videos and other sources into ready-to-use skill packs for Claude, Gemini, OpenAI and RAG frameworks like LangChain. Detects conflicts across sources, transcribes video, and exports to 21 formats.