repo_name stringlengths 2 26 | repo_link stringlengths 28 60 | category stringlengths 3 39 ⌀ | github_about_section stringlengths 22 415 | homepage_link stringlengths 14 89 ⌀ |
|---|---|---|---|---|
FTorch | https://github.com/Cambridge-ICCS/FTorch | middleware | A library for directly calling PyTorch ML models from Fortran. | https://cambridge-iccs.github.io/FTorch |
KernelBench | https://github.com/ScalingIntelligence/KernelBench | benchmark | KernelBench: Can LLMs Write GPU Kernels? - Benchmark with Torch -> CUDA problems | https://scalingintelligence.stanford.edu/blogs/kernelbench |
nvshmem | https://github.com/NVIDIA/nvshmem | distributed computing | NVIDIA NVSHMEM is a parallel programming interface for NVIDIA GPUs based on OpenSHMEM. NVSHMEM can significantly reduce multi-process communication and coordination overheads by allowing programmers to perform one-sided communication from within CUDA kernels and on CUDA streams. | https://docs.nvidia.com/nvshmem/api/index.html |
flashinfer-bench | https://github.com/flashinfer-ai/flashinfer-bench | benchmark | Building the Virtuous Cycle for AI-driven LLM Systems | https://bench.flashinfer.ai |
Primus-Turbo | https://github.com/AMD-AGI/Primus-Turbo | training framework | Primus-Turbo is a high-performance acceleration library dedicated to large-scale model training on AMD GPUs. Built and optimized for the AMD ROCm platform, it covers the full training stack — including core compute operators (GEMM, Attention, GroupedGEMM), communication primitives, optimizer modules, low-precision comp... | null |
BitBLAS | https://github.com/microsoft/BitBLAS | Basic Linear Algebra Subprograms (BLAS) | BitBLAS is a library to support mixed-precision matrix multiplications, especially for quantized LLM deployment. | null |
Wan2.2 | https://github.com/Wan-Video/Wan2.2 | video generation | Wan: Open and Advanced Large-Scale Video Generative Models | https://wan.video |
kernels-community | https://github.com/huggingface/kernels-community | gpu kernels | Kernel sources for https://huggingface.co/kernels-community | https://huggingface.co/kernels-community |
omnitrace | https://github.com/ROCm/omnitrace | performance testing | Omnitrace: Application Profiling, Tracing, and Analysis | https://rocm.docs.amd.com/projects/omnitrace |
synthetic-data-kit | https://github.com/meta-llama/synthetic-data-kit | synthetic data generation | Tool for generating high quality Synthetic datasets | https://pypi.org/project/synthetic-data-kit |
cudnn-frontend | https://github.com/NVIDIA/cudnn-frontend | parallel computing | cudnn_frontend provides a c++ wrapper for the cudnn backend API and samples on how to use it | https://developer.nvidia.com/cudnn |
PipelineRL | https://github.com/ServiceNow/PipelineRL | reinforcement learning | A scalable asynchronous reinforcement learning implementation with in-flight weight updates. | https://arxiv.org/abs/2509.19128 |
cosmos-predict2.5 | https://github.com/nvidia-cosmos/cosmos-predict2.5 | world model | Cosmos-Predict2.5, the latest version of the Cosmos World Foundation Models (WFMs) family, specialized for simulating and predicting the future state of the world in the form of video. | https://research.nvidia.com/labs/cosmos-lab/cosmos-predict2.5 |
kraken | https://github.com/meta-pytorch/kraken | kernel examples | Triton-based Symmetric Memory operators and examples | null |
TileIR | https://github.com/microsoft/TileIR | parallel computing dsl | TileIR (tile-ir) is a concise domain-specific IR designed to streamline the development of high-performance GPU/CPU kernels (e.g., GEMM, Dequant GEMM, FlashAttention, LinearAttention). By employing a Pythonic syntax with an underlying compiler infrastructure on top of TVM, TileIR allows developers to focus on productiv... | null |
intelliperf | https://github.com/AMDResearch/intelliperf | performance testing | Automated bottleneck detection and solution orchestration | https://arxiv.org/html/2508.20258v1 |
streamv2v | https://github.com/Jeff-LiangF/streamv2v | video generation | Official Pytorch implementation of StreamV2V. | https://jeff-liangf.github.io/projects/streamv2v |
tilus | https://github.com/NVIDIA/tilus | parallel computing | Tilus is a tile-level kernel programming language with explicit control over shared memory and registers. | https://nvidia.github.io/tilus |
gemlite | https://github.com/dropbox/gemlite | gpu kernels | Fast low-bit matmul kernels in Triton | null |
Self-Forcing | https://github.com/guandeh17/Self-Forcing | video generation | Official codebase for "Self Forcing: Bridging Training and Inference in Autoregressive Video Diffusion" (NeurIPS 2025 Spotlight) | https://self-forcing.github.io |
TritonBench | https://github.com/thunlp/TritonBench | benchmark | TritonBench: Benchmarking Large Language Model Capabilities for Generating Triton Operators | https://arxiv.org/abs/2502.14752 |
IMO2025 | https://github.com/harmonic-ai/IMO2025 | formal mathematical reasoning | Harmonic's model Aristotle achieved gold medal performance, solving 5 problems. This repository contains the lean statement files and proofs for Problems 1-5. | https://harmonic.fun |
RaBitQ | https://github.com/gaoj0017/RaBitQ | quantization | [SIGMOD 2024] RaBitQ: Quantizing High-Dimensional Vectors with a Theoretical Error Bound for Approximate Nearest Neighbor Search | https://github.com/VectorDB-NTU/RaBitQ-Library |
torchdendrite | https://github.com/sandialabs/torchdendrite | machine learning framework | Dendrites for PyTorch and SNNTorch neural networks | null |
triton-runner | https://github.com/toyaix/triton-runner | debugger | Multi-Level Triton Runner supporting Python, IR, PTX, and cubin. | https://triton-runner.org |
distributed-training-guide | https://github.com/LambdaLabsML/distributed-training-guide | null | Best practices & guides on how to write distributed pytorch training code | null |
Megatron-LM | https://github.com/NVIDIA/Megatron-LM | null | Ongoing research training transformer models at scale | https://docs.nvidia.com/megatron-core/developer-guide/latest/index.html |
fairscale | https://github.com/facebookresearch/fairscale | null | PyTorch extensions for high performance and large scale training. | null |
ColossalAI | https://github.com/hpcaitech/ColossalAI | null | Making large AI models cheaper, faster and more accessible | https://colossalai.org/ |
NeMo RL | https://github.com/nvidia-nemo/rl | null | Scalable toolkit for efficient model reinforcement | https://docs.nvidia.com/nemo/rl/latest/index.html |
slime | https://github.com/THUDM/slime | null | slime is an LLM post-training framework for RL Scaling. | https://thudm.github.io/slime/ |
RAGEN | https://github.com/mll-lab-nu/RAGEN | null | RAGEN leverages reinforcement learning to train LLM reasoning agents in interactive, stochastic environments. | https://ragen-ai.github.io/ |
slurm | https://github.com/SchedMD/slurm | null | Slurm: A Highly Scalable Workload Manager | https://slurm.schedmd.com/ |
Spurious Rewards | https://github.com/ruixin31/Spurious_Rewards | null | Spurious Rewards: Rethinking Training Signals in RLVR | https://arxiv.org/pdf/2506.10947 |
Qwen Code | https://github.com/QwenLM/qwen-code | null | An open-source AI coding agent that lives in your terminal. | https://qwen.ai/qwencode |
Open Thoughts | https://github.com/open-thoughts/open-thoughts | null | Fully open data curation for reasoning models | https://www.open-thoughts.ai/ |
OLMOS | https://github.com/allenai/olmes | null | Reproducible, flexible LLM evaluations | null |
SmolLM | https://github.com/huggingface/smollm | null | Everything about the SmolLM and SmolVLM family of models | null |
smolagents | https://github.com/huggingface/smolagents | null | smolagents: a barebones library for agents that think in code. | https://huggingface.co/docs/smolagents/ |
Delta Learning | https://github.com/scottgeng00/delta_learning | null | Code release for the paper "The Delta Learning Hypothesis: Preference Tuning on Weak Data can Yield Strong Gains" | https://arxiv.org/pdf/2507.06187 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.