Tom Pollak

jjui. jjui is a TUI designed for interacting with the Jujutsu version control system.

lazyjj. TUI for Jujutsu/jj

ctrl2esc. Like caps2esc, but for Chrome OS

pu. Shell

pyptx. A Python DSL to write Nvidia PTX for Hopper and Blackwell in JAX and PyTorch

sky. Sky — an Elm-inspired language that compiles to Go. Hindley-Milner types, server-driven UI (Sky.Live), single binary output.

tensor-layouts. A pure-Python implementation of the Nvidia CuTe layout algebra intended to be approachable and easy to learn.

nca-pre-pretraining. Python

autoresearch. AI agents running research on single-GPU nanochat training automatically

cli. Google Workspace CLI — one command-line tool for Drive, Gmail, Calendar, Sheets, Docs, Chat, Admin, and more. Dynamically built from Google Discovery Service. Includes AI agent skills.

discover. Python

reader3. Quick illustration of how one can easily read books together with LLMs. It's great and I highly recommend it.

lovely-tensors. Tensors, for human consumption

ibis. the portable Python dataframe library

beads. Beads - A memory upgrade for your coding agent

avataRL. rl from zero pretrain, can it be done? yes.

immer. Postmodern immutable and persistent data structures for C++ — value semantics at scale

TileGym. Helpful kernel tutorials, examples and SKILLs for tile-based GPU programming

lazygit-vscode. https://marketplace.visualstudio.com/items?itemName=TomPollak.lazygit-vscode

mini-sglang. A compact implementation of SGLang, designed to demystify the complexities of modern LLM serving systems.

nmoe. MoE training for Me and You and maybe other people

fff. The fastest and the most accurate file search SDK for AI agents, Neovim, Rust, C, Python, Bun and NodeJS

compression. llms can learn their own context compression via RL

the-art-of-debugging. The Art of Debugging Open Book

Alpha-MoE. Cuda

viztracer. A debugging and profiling tool that can trace and visualize python code execution

reference-kernels. Official Problem Sets / Reference Kernels for the GPU MODE Leaderboard!

small-malaysian-lm-B200. We want to compare how good Qwen3-1.7B-Base using B200 to continue pretraining on Malaysian multi-lingual corpus on different mixed precision training with proper truncated multi-packing.

kimi-cli. Kimi Code CLI is your next CLI agent.

rlm-minimal. Super basic implementation (gist-like) of RLMs with REPL environments.

kraken. Triton-based Symmetric Memory operators and examples

diffusion-gpt. From baby GPT to diffusion GPT: An annotated implementation of a character-level discrete diffusion model (adapted from Karpathy’s baby GPT).

TinyRecursiveModels. Python

mlir-beginner-friendly-tutorial. This is a beginner-friendly tutorial on MLIR from the perspective of a user of MLIR, not a compiler engineer. This tutorial will introduce why MLIR exists and how it is used to compile code at different levels of abstraction. This tutorial will focus on working with the "core" dialects of MLIR.

tccl. extensible collectives library in triton

openbridge. Use Claude Code with any LLM provider - GLM-4.5, Kimi-K2, Qwen3-Coder, DeepSeek, etc.

factorio-learning-environment. A non-saturating, open-ended environment for evaluating LLMs in Factorio

py-spy. Sampling profiler for Python programs

fast.cu. Fastest kernels written from scratch

quack. A Quirky Assortment of CuTe Kernels

harmony. Renderer for the harmony response format to be used with gpt-oss

HRM. Hierarchical Reasoning Model Official Release

Absolute-Zero-Reasoner. Official Repository of Absolute Zero Reasoner

atropos. Atropos is a Language Model Reinforcement Learning Environments framework for collecting and evaluating LLM trajectories through diverse environments

TorchLeet. Leetcode for Pytorch

triton-index. Cataloging released Triton kernels.

TritonAcademy. A repository to unravel the language of GPUs, making their kernel conversations easy to understand

conch. A "standard library" of Triton kernels.

Awesome-Triton-Kernels. Collection of kernels written in Triton language

memvid. Memory layer for AI Agents. Replace complex RAG pipelines with a serverless, single-file memory layer. Give your agents instant retrieval and long-term memory.

circuit-tracer. Python

ParScale. Parallel Scaling Law for Language Model — Beyond Parameter and Inference Time Scaling

Scrapling. 🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl!

toolkami. Simple Agents Made Easy

PocketFlow. Pocket Flow: 100-line LLM framework. Let Agents build Agents!

PocketFlow-Tutorial-Codebase-Knowledge. Pocket Flow: Codebase to Tutorial

prime-rl. Agentic RL Training at Scale

claude-code-proxy. Run Claude Code on OpenAI models

model2vec. Fast State-of-the-Art Static Embeddings

elimination_game. A multi-player tournament benchmark that tests LLMs in social reasoning, strategy, and deception. Players engage in public and private conversations, form alliances, and vote to eliminate each other

ptx-tutorial-by-aislop. PTX-Tutorial Written Purely By AIs (Deep Research of Openai and Claude 3.7)

machinegun.nvim. A plugin to manage multicomputer/user neovim configuration

result. A simple Rust like Result type for Python 3. Fully type annotated.

curl_cffi. Python binding for curl-impersonate fork via cffi. A http client that can impersonate browser tls/ja3/http2 fingerprints.

AeroSpace. AeroSpace is an i3-like tiling window manager for macOS

TripoSG. TripoSG: High-Fidelity 3D Shape Synthesis using Large-Scale Rectified Flow Models

nnsight. The nnsight package enables interpreting and manipulating the internals of deep learned models.

smartfunc. Turn docstrings into LLM-functions

Isaac-GR00T. NVIDIA Isaac GR00T N1.7 - A Foundation Model for Generalist Robots.

Hunyuan3D-2. High-Resolution 3D Assets Generation with Large Scale Hunyuan3D Diffusion Models.

einx. Universal Notation for Tensor Operations in Python

jsonargparse. Minimal effort CLIs derived from type hints and parse from command line, config files and environment variables

defuddle-cli. Moved to the main defuddle repo. Command line utility to extract clean html, markdown and metadata from web pages.

puffer-phc. Simplified Perpetual Humanoid Control with Pufferlib, CARBS

understand-r1-zero. Understanding R1-Zero-Like Training: A Critical Perspective

fastopenapi. FastOpenAPI is a library for generating and integrating OpenAPI schemas using Pydantic v2 and various frameworks (AioHttp, Django, Falcon, Flask, Quart, Sanic, Starlette, Tornado).

system-prompts-and-models-of-ai-tools. FULL Augment Code, Claude Code, Cluely, CodeBuddy, Comet, Cursor, Devin AI, Junie, Kiro, Leap.new, Lovable, Manus, NotionAI, Orchids.app, Perplexity, Poke, Qoder, Replit, Same.dev, Trae, Traycer AI, VSCode Agent, Warp.dev, Windsurf, Xcode, Z.ai Code, Dia & v0. (And other Open Sourced) System Prompts, Internal Tools & AI Models

verifiers. Our library for RL environments + evals

data-oriented-design. A curated list of data oriented design resources.

superfile. Pretty fancy and modern terminal file manager

structlog. Simple, powerful, and fast logging for Python.

stamina. Production-grade retries for Python

OpenDiloco. OpenDiLoCo: An Open-Source Framework for Globally Distributed Low-Communication Training

s1. s1: Simple test-time scaling

snacks.nvim. 🍿 A collection of QoL plugins for Neovim

TinyZero. Minimal reproduction of DeepSeek R1-Zero

person_search. Joint Detection and Identification Feature Learning for Person Search

RL. Python

InstantMesh. InstantMesh: Efficient 3D Mesh Generation from a Single Image with Sparse-view Large Reconstruction Models

minimind. 🧠「大模型」2小时完全从0训练64M的小参数LLM！Train a 64M-parameter LLM from scratch in just 2h!

Hunyuan3D-1. Tencent Hunyuan3D-1.0: A Unified Framework for Text-to-3D and Image-to-3D Generation

LGM. [ECCV 2024 Oral] LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation.

Wonder3D. Single Image to 3D using Cross-Domain Diffusion for 3D Generation

d2l-en. Interactive deep learning book with multi-framework code, math, and discussions. Adopted at 500 universities from 70 countries including Stanford, MIT, Harvard, and Cambridge.

B-LoRA. Implicit Style-Content Separation using B-LoRA

stable-fast-3d. SF3D: Stable Fast 3D Mesh Reconstruction with UV-unwrapping and Illumination Disentanglement

blender-projects. Find Game assets, loops, environment materials, and shaders created using the open-source blender ❤ toolkit.

blink.cmp. Performant, batteries-included completion plugin for Neovim

concentration_notebooks. Jupyter Notebook

Monet. [ICLR 2025] Monet: Mixture of Monosemantic Experts for Transformers

tiny-activation-dashboard. A tiny easily hackable implementation of a feature dashboard.

IC-Light. More relighting!

crosscoder_learning. Modified to support crosscoder training.

eufy-robovac. NodeJS library to control Eufy RoboVac

alphazero-general. A fast, generalized, and modified implementation of Deepmind's distinguished AlphaZero in PyTorch.

reid-strong-baseline. Bag of Tricks and A Strong Baseline for Deep Person Re-identification

SAELens. Training Sparse Autoencoders on Language Models

imagenet.int8. Python

lingua. Meta Lingua: a lean, efficient, and easy-to-hack codebase to research LLMs.

vitsae. Jupyter Notebook

sd3.5. Python

shortcut-models. Python

MishformerLens. MishformerLens intends to be a drop-in replacement for TransformerLens that AST patches HuggingFace Transformers rather than implementing a custom, numerically inaccurate Transformer architecture.

bm25_pt. minimal pytorch implementation of bm25 (with sparse tensors)

cosmopolitan. build-once run-anywhere c library

clip-distillation. Zero-label image classification via OpenCLIP knowledge distillation

supervision. We write your reusable computer vision tools. 💜

MobileQuant. [EMNLP Findings 2024] MobileQuant: Mobile-friendly Quantization for On-device Language Models

pydantic_config. Manage ML configuration with pydantic

x-flux-comfyui. Python

loom. A Loom implementation in Obsidian

ghapi. A delightful and complete interface to GitHub's amazing API

MLE-agent. 🤖 MLE-Agent: Your intelligent companion for seamless AI engineering and research. 🔍 Integrate with arxiv and paper with code to provide better code/research plans 🧰 OpenAI, Anthropic, Gemini, Ollama, etc supported. :fireworks: Code RAG

elia. A snappy, keyboard-centric terminal user interface for interacting with large language models. Chat with ChatGPT, Claude, Llama 3, Phi 3, Mistral, Gemma and more.

humanify. Deobfuscate Javascript code using ChatGPT

AnkiBrain. Python

build-your-own-x. Master programming by recreating your favorite technologies from scratch.

hide. 🤖 Headless IDE for AI agents

firecrawl. The API to search, scrape, and interact with the web at scale. 🔥

Show-o. [ICLR & NeurIPS 2025] Repository for Show-o series, One Single Transformer to Unify Multimodal Understanding and Generation.

Liger-Kernel. Efficient Triton Kernels for LLM Training

mcts-llm. Python

ADAS. [ICLR 2025] Automated Design of Agentic Systems

hailo_model_zoo. The Hailo Model Zoo includes pre-trained models and a full building and evaluation environment

ai-toolkit. The ultimate training toolkit for finetuning diffusion models

nanotron. Minimalistic large language model 3D-parallelism training

basedpyright. pyright fork with various type checking improvements, improved vscode support and pylance features built into the language server

textual. The lean application framework for Python. Build sophisticated user interfaces with a simple Python API. Run your apps in the terminal and a web browser.

Prompt-Engineering-Guide. 🐙 Guides, papers, lessons, notebooks and resources for prompt engineering, context engineering, RAG, and AI Agents.

PufferLib. Puffing up reinforcement learning

delphi. Delphi was the home of a temple to Phoebus Apollo, which famously had the inscription, 'Know Thyself.' This library lets language models know themselves through automated interpretability.

ViCHA. [BMVC22] Official Implementation of ViCHA: "Efficient Vision-Language Pretraining with Visual Concepts and Hierarchical Alignment"

aider. aider is AI pair programming in your terminal

TextbookGPT. The ChatGPT Textbook Plugin lets you easily search information in a textbook by asking questions in everyday language.

txtai. 💡 All-in-one AI framework for semantic search, LLM orchestration and language model workflows

duckdb-embedding-search. Fast similarity search using DuckDB

rr. Record and Replay Framework

sanitizers. AddressSanitizer, ThreadSanitizer, MemorySanitizer

TinyLLaVA_Factory. A Framework of Small-scale Large Multimodal Models

ziglings. Extensive introductory writeup on Zig language functionalities

ComfyUI. The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.

sparsify. Sparsify transformers with SAEs and transcoders

r-nn. Tensor library with autograd using only Rust's standard library

clip-jax. Train vision models using JAX and 🤗 transformers

GPT-2. Following Karpathy with GPT-2 implementation and training, writing lots of comments cause I have memory of a goldfish

MeshAnything. [ICLR 2025] From anything to mesh like human artists. Official impl. of "MeshAnything: Artist-Created Mesh Generation with Autoregressive Transformers"

neovim. Vim-fork focused on extensibility and usability

maestro. A framework for Claude Opus to intelligently orchestrate subagents.

claude-engineer. Claude Engineer is an interactive command-line interface (CLI) that leverages the power of Anthropic's Claude-3.5-Sonnet model to assist with software development tasks.This framework enables Claude to generate and manage its own tools, continuously expanding its capabilities through conversation. Available both as a CLI and a modern web interface

audiocraft. Audiocraft is a library for audio processing and generation with deep learning. It features the state-of-the-art EnCodec audio compressor / tokenizer, along with MusicGen, a simple and controllable music generation LM with textual and melodic conditioning.

ml-4m. 4M: Massively Multimodal Masked Modeling

arcprize. Python

LLM101n. LLM101n: Let's build a Storyteller

MathBlackBox. Python

clip-text-directions. Jupyter Notebook

gpt-fast. Simple and efficient pytorch-native transformer text generation in <1000 LOC of python.

AutoLoRADiscovery. Jupyter Notebook

Vehicle_reID-Collection. :red_car: the collection of vehicle re-ID papers, datasets. :red_car:

c-style. My favorite C programming practices.

ftcourse. Jupyter Notebook

lora_instruction_tune. Python

dspy. DSPy: The framework for programming—not prompting—language models

llama3-from-scratch. llama3 implementation one matrix multiplication at a time

ThunderKittens. Tile primitives for speedy kernels

min-max-gpt. Minimal (400 LOC) implementation Maximum (multi-node, FSDP) GPT training

Glider. Open-source E-ink monitor. Mirror of https://gitlab.com/zephray/glider

SQLiteCpp. SQLiteC++ (SQLiteCpp) is a smart and easy to use C++ SQLite3 wrapper.

interface99. Full-featured interfaces for C99

datatype99. Algebraic data types for C99

clip-decomposition. Jupyter Notebook

unsloth. Unsloth is a local UI for training and running Kimi K3, Gemma 4, Qwen3.6, DeepSeek, GLM and other models.

quix-streams. Python Streaming DataFrames for Kafka

clipora. Clipora is a powerful toolkit for fine-tuning OpenCLIP models using Low Rank Adapters (LoRA).

pyreft. Stanford NLP Python library for Representation Finetuning (ReFT)

machine-learning-list. A curriculum for learning about foundation models, from scratch to the frontier

convmixer. Implementation of ConvMixer for "Patches Are All You Need? 🤷"

Transformer-Puzzles. Puzzles for exploring transformers

resource-stream. GPU programming related news and material links

chess_llm_interpretability. Visualizing the internal board state of a GPT trained on chess PGN strings, and performing interventions on its internal board state and representation of player Elo.

lumentis. AI powered one-click comprehensive docs from transcripts and text.

files-to-prompt. Concatenate a directory full of files into a single prompt for use with LLMs

flash-attention-minimal. Flash Attention in ~100 lines of CUDA (forward pass only)

LlamaGym. Fine-tune LLM agents with online reinforcement learning

prql. PRQL is a modern language for transforming data — a simple, powerful, pipelined SQL replacement

ARC-AGI. The Abstraction and Reasoning Corpus

cv. Print-friendly, minimalist CV page

Bunny. A family of lightweight multimodal models.

SMILEtrack. Jupyter Notebook

vscode-paste-image. paste image from clipboard to markdown/asciidoc directly!

ml-engineering. Machine Learning Engineering Open Book