Dev
GitHub repos gaining traction - what high-signal users are starring and what's climbing the board, captured daily and enriched from GitHub. Raw material for spotting new tech and patterns worth building on.
3,195
repos tracked
105
surfaced this week
193
created < 30d
Python
top language
101 repos
-
DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.
-
an rlm-style training harness
-
Machine Learning Engineering Open Book
-
Cheap character training: LoRA adapters trained on Base, grafted onto Instruct. The character is Musonius, from a virtue-ethics constitution.
-
nnScaler: Compiling DNN models for Parallel Training
-
Implement a ChatGPT-like LLM in PyTorch from scratch, step by step
-
Agentic RL Training at Scale
-
AllenAI's post-training codebase
-
This module defines a type system for distributed training code, based off of JAX's sharding in types, but adapted for the PyTorch ecosystem.
-
Multimodal RL training framework for diffusion & omni models
-
A toy eval suite for tracing generalization dynamics of LM pre-training
-
Programmable chat templates for LLM training and inference.
-
Official CLI and Python SDK for Prime Intellect - access GPU compute, remote sandboxes, RL environments, and distributed training infrastructure for AI development at scale.
-
Executable RL environment for diagnosing and repairing distributed PyTorch training failures, with sandboxed execution and correctness-verified recovery rewards.
-
Speedrun for pretraining without backpropagation
-
Training Code to endow language models with a sense of proprioception
-
An open-source document parsing model with SOTA accuracy and speed, using an RSI loop during post-training.
-
The largest collection of PyTorch image encoders / backbones. Including train, eval, inference, export scripts, and pretrained weights -- ResNet, ResNeXT, EfficientNet, NFNet, Vision Transformer (ViT), MobileNetV4, MobileNet-V3 & V2, RegNet, DPN, CSPNet, Swin Transformer, MaxViT, CoAtNet, ConvNeXt, and more
-
Compile natural-language function descriptions into reusable local PAW programs.
-
An interface library for RL post training with environments.
-
Hexapod walker: STS3215 prototype, RL training orchestrator (CoreWeave), MuJoCo sim, robot control
-
Feather agent harness, training integrations, and evaluation tools for knowledge work
-
Official implementation of "Correction Training for Diffusion Models"
-
An interface library for RL post training with environments.
-
Fast, flexible, elastically deterministic data loading for foundation model training
-
Quantifying religious content in LLM pretraining data — corpus analysis measuring Christian and other religious traditions in open-source training datasets
-
🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.
-
Post-training with Tinker
-
A probe for the 'training intuition' of LLMs (and human).
-
Training Jev-like decision models using Qwen3 with choice-order invariance
-
Loom TUI and Character Lab for developing AI characters from base models: branch text, curate anthologies, simulate conversations, and evaluate training data.
-
RL training environments for Microduck (mjlab)
-
Halo is an open-source framework built by White Circle for training large language and multimodal models
-
verl/HybridFlow: A Flexible and Efficient RL Post-Training Framework
-
A curated collection of papers, technical reports, frameworks, and tools for on-policy distillation (OPD) of large language models
-
Full-weight Qwen decision model with calibrated probabilities, training code, REST API, and playground.
-
Fast, versioned weight updates for LLM serving and training
-
Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.
-
a toolkit for training privileged humanoid controllers
-
Train speculative decoding models effortlessly and port them smoothly to SGLang serving.
-
NVIDIA Resiliency Extension is a python package for framework developers and users to implement fault-tolerant features. It improves the effective training time by minimizing the downtime due to failures and interruptions.
-
Training Large Language Model to Reason in a Continuous Latent Space
-
Infrastructure for continually self‑improving agents
-
End-to-end project training and building AI models / LLMs to detect AI-written text
-
The AI Compute Platform for frontier teams. SkyPilot turns fragmented AI compute into one AI supercomputer, so frontier AI teams build custom intelligence faster.
-
RL training environments for Microduck (mjlab)
-
Game Env for RL training/self-play/curriculum learning