[
  {
    "repo_name": "llvm-project",
    "repo_link": "https://github.com/llvm/llvm-project",
    "category": "compiler",
    "github_about_section": "The LLVM Project is a collection of modular and reusable compiler and toolchain technologies.",
    "homepage_link": "http://llvm.org"
  },
  {
    "repo_name": "pytorch",
    "repo_link": "https://github.com/pytorch/pytorch",
    "category": "machine learning framework",
    "github_about_section": "Tensors and Dynamic neural networks in Python with strong GPU acceleration",
    "homepage_link": "https://pytorch.org"
  },
  {
    "repo_name": "kubernetes",
    "repo_link": "https://github.com/kubernetes/kubernetes",
    "category": "container orchestration",
    "github_about_section": "Production-Grade Container Scheduling and Management",
    "homepage_link": "https://kubernetes.io"
  },
  {
    "repo_name": "tensorflow",
    "repo_link": "https://github.com/tensorflow/tensorflow",
    "category": "machine learning framework",
    "github_about_section": "An Open Source Machine Learning Framework for Everyone",
    "homepage_link": "https://tensorflow.org"
  },
  {
    "repo_name": "transformers",
    "repo_link": "https://github.com/huggingface/transformers",
    "category": "multi-purpose library",
    "github_about_section": "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.",
    "homepage_link": "https://huggingface.co/transformers"
  },
  {
    "repo_name": "spark",
    "repo_link": "https://github.com/apache/spark",
    "category": "data processing",
    "github_about_section": "Apache Spark - A unified analytics engine for large-scale data processing",
    "homepage_link": "https://spark.apache.org"
  },
  {
    "repo_name": "hhvm",
    "repo_link": "https://github.com/facebook/hhvm",
    "category": "virtual machine",
    "github_about_section": "A virtual machine for executing programs written in Hack.",
    "homepage_link": "https://hhvm.com"
  },
  {
    "repo_name": "vllm",
    "repo_link": "https://github.com/vllm-project/vllm",
    "category": "inference engine",
    "github_about_section": "A high-throughput and memory-efficient inference and serving engine for LLMs",
    "homepage_link": "https://docs.vllm.ai"
  },
  {
    "repo_name": "elasticsearch",
    "repo_link": "https://github.com/elastic/elasticsearch",
    "category": "search engine",
    "github_about_section": "Free and Open Source, Distributed, RESTful Search Engine",
    "homepage_link": "https://elastic.co/products/elasticsearch"
  },
  {
    "repo_name": "numpy",
    "repo_link": "https://github.com/numpy/numpy",
    "category": "scientific computing",
    "github_about_section": "The fundamental package for scientific computing with Python.",
    "homepage_link": "https://numpy.org"
  },
  {
    "repo_name": "scipy",
    "repo_link": "https://github.com/scipy/scipy",
    "category": "scientific computing",
    "github_about_section": "SciPy library main repository",
    "homepage_link": "https://scipy.org"
  },
  {
    "repo_name": "llama.cpp",
    "repo_link": "https://github.com/ggml-org/llama.cpp",
    "category": "inference engine",
    "github_about_section": "LLM inference in C/C++",
    "homepage_link": "https://ggml.ai"
  },
  {
    "repo_name": "ray",
    "repo_link": "https://github.com/ray-project/ray",
    "category": "multi-purpose library",
    "github_about_section": "Ray is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.",
    "homepage_link": "https://ray.io"
  },
  {
    "repo_name": "sglang",
    "repo_link": "https://github.com/sgl-project/sglang",
    "category": "inference engine",
    "github_about_section": "SGLang is a fast serving framework for large language models and vision language models.",
    "homepage_link": "https://docs.sglang.ai"
  },
  {
    "repo_name": "rocm-systems",
    "repo_link": "https://github.com/ROCm/rocm-systems",
    "category": "multi-purpose library",
    "github_about_section": "super repo for rocm systems projects",
    "homepage_link": "https://amd.com/en/products/software/rocm.html"
  },
  {
    "repo_name": "jax",
    "repo_link": "https://github.com/jax-ml/jax",
    "category": "scientific computing",
    "github_about_section": "Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more",
    "homepage_link": "https://docs.jax.dev"
  },
  {
    "repo_name": "onnxruntime",
    "repo_link": "https://github.com/microsoft/onnxruntime",
    "category": "machine learning interoperability",
    "github_about_section": "ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator",
    "homepage_link": "https://onnxruntime.ai"
  },
  {
    "repo_name": "verl",
    "repo_link": "https://github.com/volcengine/verl",
    "category": "reinforcement learning",
    "github_about_section": "verl: Volcano Engine Reinforcement Learning for LLMs",
    "homepage_link": "https://verl.readthedocs.io"
  },
  {
    "repo_name": "ollama",
    "repo_link": "https://github.com/ollama/ollama",
    "category": "inference engine",
    "github_about_section": "Get up and running with OpenAI gpt-oss, DeepSeek-R1, Gemma 3 and other models.",
    "homepage_link": "https://ollama.com"
  },
  {
    "repo_name": "triton",
    "repo_link": "https://github.com/triton-lang/triton",
    "category": "parallel computing dsl",
    "github_about_section": "Development repository for the Triton language and compiler",
    "homepage_link": "https://triton-lang.org"
  },
  {
    "repo_name": "executorch",
    "repo_link": "https://github.com/pytorch/executorch",
    "category": "model compiler",
    "github_about_section": "On-device AI across mobile, embedded and edge for PyTorch",
    "homepage_link": "https://executorch.ai"
  },
  {
    "repo_name": "triton-windows",
    "repo_link": "https://github.com/woct0rdho/triton-windows",
    "category": "parallel computing dsl",
    "github_about_section": "Fork of the Triton language and compiler for Windows support and easy installation"
  },
  {
    "repo_name": "trl",
    "repo_link": "https://github.com/huggingface/trl",
    "category": "reinforcement learning",
    "github_about_section": "Train transformer language models with reinforcement learning.",
    "homepage_link": "http://hf.co/docs/trl"
  },
  {
    "repo_name": "goose",
    "repo_link": "https://github.com/block/goose",
    "category": "agent",
    "github_about_section": "an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM",
    "homepage_link": "https://block.github.io/goose"
  },
  {
    "repo_name": "DeepSpeed",
    "repo_link": "https://github.com/deepspeedai/DeepSpeed",
    "category": "training framework",
    "github_about_section": "DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.",
    "homepage_link": "https://deepspeed.ai"
  },
  {
    "repo_name": "numba",
    "repo_link": "https://github.com/numba/numba",
    "category": "compiler",
    "github_about_section": "NumPy aware dynamic Python compiler using LLVM",
    "homepage_link": "https://numba.pydata.org"
  },
  {
    "repo_name": "torchrec",
    "repo_link": "https://github.com/meta-pytorch/torchrec",
    "category": "distributed computing",
    "github_about_section": "Pytorch domain library for recommendation systems",
    "homepage_link": "https://meta-pytorch.org/torchrec/"
  },
  {
    "repo_name": "modular",
    "repo_link": "https://github.com/modular/modular",
    "category": "parallel computing",
    "github_about_section": "The Modular Platform (includes MAX & Mojo)",
    "homepage_link": "https://docs.modular.com"
  },
  {
    "repo_name": "accelerate",
    "repo_link": "https://github.com/huggingface/accelerate",
    "category": "training framework",
    "github_about_section": "A simple way to launch, train, and use PyTorch models on almost any device and distributed configuration, automatic mixed precision (including fp8), and easy-to-configure FSDP and DeepSpeed support.",
    "homepage_link": "https://huggingface.co/docs/accelerate"
  },
  {
    "repo_name": "milvus",
    "repo_link": "https://github.com/milvus-io/milvus",
    "category": "vector database",
    "github_about_section": "Milvus is a high-performance, cloud-native vector database built for scalable vector ANN search",
    "homepage_link": "https://milvus.io"
  },
  {
    "repo_name": "onnx",
    "repo_link": "https://github.com/onnx/onnx",
    "category": "machine learning interoperability",
    "github_about_section": "Open standard for machine learning interoperability",
    "homepage_link": "https://onnx.ai"
  },
  {
    "repo_name": "modelcontextprotocol",
    "repo_link": "https://github.com/modelcontextprotocol/modelcontextprotocol",
    "category": "mcp",
    "github_about_section": "Specification and documentation for the Model Context Protocol",
    "homepage_link": "https://modelcontextprotocol.io"
  },
  {
    "repo_name": "ComfyUI",
    "repo_link": "https://github.com/comfyanonymous/ComfyUI",
    "category": "user interface",
    "github_about_section": "The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.",
    "homepage_link": "https://comfy.org"
  },
  {
    "repo_name": "peft",
    "repo_link": "https://github.com/huggingface/peft",
    "category": "fine tuning",
    "github_about_section": "PEFT: State-of-the-art Parameter-Efficient Fine-Tuning.",
    "homepage_link": "https://huggingface.co/docs/peft"
  },
  {
    "repo_name": "flashinfer",
    "repo_link": "https://github.com/flashinfer-ai/flashinfer",
    "category": "gpu kernels",
    "github_about_section": "FlashInfer: Kernel Library for LLM Serving",
    "homepage_link": "https://flashinfer.ai"
  },
  {
    "repo_name": "trainer",
    "repo_link": "https://github.com/kubeflow/trainer",
    "github_about_section": "Distributed AI Model Training and LLM Fine-Tuning on Kubernetes",
    "homepage_link": "https://www.kubeflow.org/docs/components/trainer/"
  },
  {
    "repo_name": "cutlass",
    "repo_link": "https://github.com/NVIDIA/cutlass",
    "category": "parallel computing",
    "github_about_section": "CUDA Templates and Python DSLs for High-Performance Linear Algebra",
    "homepage_link": "https://docs.nvidia.com/cutlass/index.html"
  },
  {
    "repo_name": "aiter",
    "repo_link": "https://github.com/ROCm/aiter",
    "category": "gpu kernels",
    "github_about_section": "AI Tensor Engine for ROCm",
    "homepage_link": "https://rocm.blogs.amd.com/software-tools-optimization/aiter-ai-tensor-engine/README.html"
  },
  {
    "repo_name": "wandb",
    "repo_link": "https://github.com/wandb/wandb",
    "category": "ml visualization",
    "github_about_section": "The AI developer platform. Use Weights & Biases to train and fine-tune models, and manage models from experimentation to production.",
    "homepage_link": "https://wandb.ai"
  },
  {
    "repo_name": "ao",
    "repo_link": "https://github.com/pytorch/ao",
    "category": "quantization",
    "github_about_section": "PyTorch native quantization and sparsity for training and inference",
    "homepage_link": "https://pytorch.org/ao"
  },
  {
    "repo_name": "Mooncake",
    "repo_link": "https://github.com/kvcache-ai/Mooncake",
    "category": "inference",
    "github_about_section": "Mooncake is the serving platform for Kimi, a leading LLM service provided by Moonshot AI.",
    "homepage_link": "https://kvcache-ai.github.io/Mooncake"
  },
  {
    "repo_name": "ccache",
    "repo_link": "https://github.com/ccache/ccache",
    "category": "compiler",
    "github_about_section": "ccache - a fast compiler cache",
    "homepage_link": "https://ccache.dev"
  },
  {
    "repo_name": "unsloth",
    "repo_link": "https://github.com/unslothai/unsloth",
    "category": "fine tuning",
    "github_about_section": "Fine-tuning & Reinforcement Learning for LLMs. Train OpenAI gpt-oss, DeepSeek-R1, Qwen3, Gemma 3, TTS 2x faster with 70% less VRAM.",
    "homepage_link": "https://docs.unsloth.ai"
  },
  {
    "repo_name": "torchtitan",
    "repo_link": "https://github.com/pytorch/torchtitan",
    "category": "training framework",
    "github_about_section": "A PyTorch native platform for training generative AI models",
    "homepage_link": "https://arxiv.org/abs/2410.06511"
  },
  {
    "repo_name": "LMCache",
    "repo_link": "https://github.com/LMCache/LMCache",
    "category": "inference",
    "github_about_section": "Supercharge Your LLM with the Fastest KV Cache Layer",
    "homepage_link": "https://lmcache.ai"
  },
  {
    "repo_name": "lapack",
    "repo_link": "https://github.com/Reference-LAPACK/lapack",
    "category": "linear algebra",
    "github_about_section": "LAPACK is a library of Fortran subroutines for solving the most commonly occurring problems in numerical linear algebra.",
    "homepage_link": "https://netlib.org/lapack"
  },
  {
    "repo_name": "FlagGems",
    "repo_link": "https://github.com/flagos-ai/FlagGems",
    "github_about_section": "FlagGems is an operator library for large language models implemented in the Triton Language."
  },
  {
    "repo_name": "ROCm",
    "repo_link": "https://github.com/ROCm/ROCm",
    "category": "multi-purpose library",
    "github_about_section": "AMD ROCm Software - GitHub Home",
    "homepage_link": "https://rocm.docs.amd.com"
  },
  {
    "repo_name": "letta",
    "repo_link": "https://github.com/letta-ai/letta",
    "category": "agent",
    "github_about_section": "Letta is the platform for building stateful agents: open AI with advanced memory that can learn and self-improve over time.",
    "homepage_link": "https://docs.letta.com"
  },
  {
    "repo_name": "server",
    "repo_link": "https://github.com/triton-inference-server/server",
    "category": "inference server",
    "github_about_section": "The Triton Inference Server provides an optimized cloud and edge inferencing solution.",
    "homepage_link": "https://docs.nvidia.com/deeplearning/triton-inference-server/user-guide/docs/index.html"
  },
  {
    "repo_name": "aws-neuron-sdk",
    "repo_link": "https://github.com/aws-neuron/aws-neuron-sdk",
    "category": "sdk",
    "github_about_section": "Powering AWS purpose-built machine learning chips. Blazing fast and cost effective, natively integrated into PyTorch and TensorFlow and integrated with your favorite AWS services",
    "homepage_link": "https://aws.amazon.com/ai/machine-learning/neuron"
  },
  {
    "repo_name": "Liger-Kernel",
    "repo_link": "https://github.com/linkedin/Liger-Kernel",
    "category": "kernel examples",
    "github_about_section": "Efficient Triton Kernels for LLM Training",
    "homepage_link": "https://openreview.net/pdf?id=36SjAIT42G"
  },
  {
    "repo_name": "tilelang",
    "repo_link": "https://github.com/tile-ai/tilelang",
    "category": "parallel computing dsl",
    "github_about_section": "Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels",
    "homepage_link": "https://tilelang.com"
  },
  {
    "repo_name": "metaflow",
    "repo_link": "https://github.com/Netflix/metaflow",
    "category": "container orchestration",
    "github_about_section": "Build, Manage and Deploy AI/ML Systems",
    "homepage_link": "https://metaflow.org"
  },
  {
    "repo_name": "litgpt",
    "repo_link": "https://github.com/Lightning-AI/litgpt",
    "github_about_section": "20+ high-performance LLMs with recipes to pretrain, finetune and deploy at scale.",
    "homepage_link": "https://lightning.ai/"
  },
  {
    "repo_name": "monarch",
    "repo_link": "https://github.com/meta-pytorch/monarch",
    "category": "distributed computing",
    "github_about_section": "PyTorch Single Controller",
    "homepage_link": "https://meta-pytorch.org/monarch"
  },
  {
    "repo_name": "hipBLASLt",
    "repo_link": "https://github.com/AMD-AGI/hipBLASLt",
    "category": "Basic Linear Algebra Subprograms (BLAS)",
    "github_about_section": "hipBLASLt is a library that provides general matrix-matrix operations with a flexible API and extends functionalities beyond a traditional BLAS library",
    "homepage_link": "https://rocm.docs.amd.com/projects/hipBLASLt"
  },
  {
    "repo_name": "flash-linear-attention",
    "repo_link": "https://github.com/fla-org/flash-linear-attention",
    "category": "gpu kernels",
    "github_about_section": "Efficient implementations of state-of-the-art linear attention models"
  },
  {
    "repo_name": "TensorRT",
    "repo_link": "https://github.com/NVIDIA/TensorRT",
    "category": "inference engine",
    "github_about_section": "NVIDIA TensorRT is an SDK for high-performance deep learning inference on NVIDIA GPUs. This repository contains the open source components of TensorRT.",
    "homepage_link": "https://developer.nvidia.com/tensorrt"
  },
  {
    "repo_name": "AReal",
    "repo_link": "https://github.com/inclusionAI/AReaL",
    "category": "reinforcement learning",
    "github_about_section": "The RL Bridge for LLM-based Agent Applications. Made Simple & Flexible.",
    "homepage_link": "https://www.inclusion-ai.org/AReaL"
  },
  {
    "repo_name": "terminal-bench",
    "repo_link": "https://github.com/laude-institute/terminal-bench",
    "category": "benchmark",
    "github_about_section": "A benchmark for LLMs on complicated tasks in the terminal",
    "homepage_link": "https://tbench.ai"
  },
  {
    "repo_name": "warp",
    "repo_link": "https://github.com/NVIDIA/warp",
    "category": "spatial computing",
    "github_about_section": "A Python framework for accelerated simulation, data generation and spatial computing.",
    "homepage_link": "https://nvidia.github.io/warp"
  },
  {
    "repo_name": "OpenRLHF",
    "repo_link": "https://github.com/OpenRLHF/OpenRLHF",
    "category": "reinforcement learning",
    "github_about_section": "An Easy-to-use, Scalable and High-performance Agentic RL Framework based on Ray (PPO & DAPO & REINFORCE++ & VLM & TIS & vLLM & Ray & Async RL)",
    "homepage_link": "https://openrlhf.readthedocs.io"
  },
  {
    "repo_name": "truss",
    "repo_link": "https://github.com/basetenlabs/truss",
    "category": "inference engine",
    "github_about_section": "The simplest way to serve AI/ML models in production",
    "homepage_link": "https://truss.baseten.co"
  },
  {
    "repo_name": "RLinf",
    "repo_link": "https://github.com/RLinf/RLinf",
    "category": "reinforcement learning",
    "github_about_section": "RLinf: Reinforcement Learning Infrastructure for Embodied and Agentic AI",
    "homepage_link": "https://rlinf.readthedocs.io"
  },
  {
    "repo_name": "nccl",
    "repo_link": "https://github.com/NVIDIA/nccl",
    "category": "distributed computing",
    "github_about_section": "Optimized primitives for collective multi-GPU communication",
    "homepage_link": "https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/index.html"
  },
  {
    "repo_name": "SkyRL",
    "repo_link": "https://github.com/NovaSky-AI/SkyRL",
    "category": "reinforcement learning",
    "github_about_section": "SkyRL: A Modular Full-stack RL Library for LLMs",
    "homepage_link": "https://docs.skyrl.ai/docs"
  },
  {
    "repo_name": "ROLL",
    "repo_link": "https://github.com/alibaba/ROLL",
    "category": "reinforcement learning",
    "github_about_section": "An Efficient and User-Friendly Scaling Library for Reinforcement Learning with Large Language Models",
    "homepage_link": "https://alibaba.github.io/ROLL/"
  },
  {
    "repo_name": "lightning-thunder",
    "repo_link": "https://github.com/Lightning-AI/lightning-thunder",
    "category": "model compiler",
    "github_about_section": "PyTorch compiler that accelerates training and inference. Get built-in optimizations for performance, memory, parallelism, and easily write your own."
  },
  {
    "repo_name": "helion",
    "repo_link": "https://github.com/pytorch/helion",
    "category": "parallel computing dsl",
    "github_about_section": "A Python-embedded DSL that makes it easy to write fast, scalable ML kernels with minimal boilerplate.",
    "homepage_link": "https://helionlang.com"
  },
  {
    "repo_name": "ort",
    "repo_link": "https://github.com/pykeio/ort",
    "category": "machine learning interoperability",
    "github_about_section": "Fast ML inference & training for ONNX models in Rust",
    "homepage_link": "https://ort.pyke.io"
  },
  {
    "repo_name": "deepinv",
    "repo_link": "https://github.com/deepinv/deepinv",
    "github_about_section": "DeepInverse: a PyTorch library for solving imaging inverse problems using deep learning",
    "homepage_link": "https://deepinv.github.io/deepinv/"
  },
  {
    "repo_name": "dstack",
    "repo_link": "https://github.com/dstackai/dstack",
    "category": "container orchestration",
    "github_about_section": "dstack is an open-source control plane for running development, training, and inference jobs on GPUs-across hyperscalers, neoclouds, or on-prem.",
    "homepage_link": "https://dstack.ai"
  },
  {
    "repo_name": "doctr",
    "repo_link": "https://github.com/mindee/doctr",
    "github_about_section": "docTR (Document Text Recognition) - a seamless, high-performing & accessible library for OCR-related tasks powered by Deep Learning.",
    "homepage_link": "https://mindee.github.io/doctr/"
  },
  {
    "repo_name": "SWE-bench",
    "repo_link": "https://github.com/SWE-bench/SWE-bench",
    "category": "benchmark",
    "github_about_section": "SWE-bench: Can Language Models Resolve Real-world Github Issues?",
    "homepage_link": "https://swebench.com"
  },
  {
    "repo_name": "torchchat",
    "repo_link": "https://github.com/pytorch/torchchat",
    "github_about_section": "Run PyTorch LLMs locally on servers, desktop and mobile"
  },
  {
    "repo_name": "mcp-agent",
    "repo_link": "https://github.com/lastmile-ai/mcp-agent",
    "category": "mcp",
    "github_about_section": "Build effective agents using Model Context Protocol and simple workflow patterns"
  },
  {
    "repo_name": "prime-rl",
    "repo_link": "https://github.com/PrimeIntellect-ai/prime-rl",
    "category": "reinforcement learning",
    "github_about_section": "Agentic RL Training at Scale"
  },
  {
    "repo_name": "cuda-python",
    "repo_link": "https://github.com/NVIDIA/cuda-python",
    "category": "middleware",
    "github_about_section": "CUDA Python: Performance meets Productivity",
    "homepage_link": "https://nvidia.github.io/cuda-python"
  },
  {
    "repo_name": "open-instruct",
    "repo_link": "https://github.com/allenai/open-instruct",
    "category": "reinforcement learning",
    "github_about_section": "AllenAI's post-training codebase",
    "homepage_link": "https://allenai.github.io/open-instruct/"
  },
  {
    "repo_name": "openevolve",
    "repo_link": "https://github.com/codelion/openevolve",
    "category": "evolutionary algorithm",
    "github_about_section": "Open-source implementation of AlphaEvolve"
  },
  {
    "repo_name": "litserve",
    "repo_link": "https://github.com/Lightning-AI/litserve",
    "github_about_section": "A minimal Python framework for building custom AI inference servers with full control over logic, batching, and scaling.",
    "homepage_link": "https://lightning.ai/litserve"
  },
  {
    "repo_name": "openzl",
    "repo_link": "https://github.com/facebook/openzl",
    "category": "data compression",
    "github_about_section": "A novel data compression framework",
    "homepage_link": "https://openzl.org"
  },
  {
    "repo_name": "torchforge",
    "repo_link": "https://github.com/meta-pytorch/torchforge",
    "category": "reinforcement learning",
    "github_about_section": "PyTorch-native post-training at scale",
    "homepage_link": "https://meta-pytorch.org/torchforge"
  },
  {
    "repo_name": "torchcodec",
    "repo_link": "https://github.com/meta-pytorch/torchcodec",
    "github_about_section": "PyTorch media decoding and encoding",
    "homepage_link": "https://meta-pytorch.org/torchcodec/stable/index.html"
  },
  {
    "repo_name": "quack",
    "repo_link": "https://github.com/Dao-AILab/quack",
    "category": "kernel examples",
    "github_about_section": "A Quirky Assortment of CuTe Kernels"
  },
  {
    "repo_name": "Triton-distributed",
    "repo_link": "https://github.com/ByteDance-Seed/Triton-distributed",
    "category": "distributed computing",
    "github_about_section": "Distributed Compiler based on Triton for Parallel Systems",
    "homepage_link": "https://triton-distributed.readthedocs.io"
  },
  {
    "repo_name": "ThunderKittens",
    "repo_link": "https://github.com/HazyResearch/ThunderKittens",
    "category": "parallel computing",
    "github_about_section": "Tile primitives for speedy kernels",
    "homepage_link": "https://hazyresearch.stanford.edu/blog/2024-10-29-tk2"
  },
  {
    "repo_name": "kernels",
    "repo_link": "https://github.com/huggingface/kernels",
    "category": "gpu kernels",
    "github_about_section": "Load compute kernels from the Hub"
  },
  {
    "repo_name": "GEAK-agent",
    "repo_link": "https://github.com/AMD-AGI/GEAK-agent",
    "category": "agent",
    "github_about_section": "It is an LLM-based AI agent, which can write correct and efficient gpu kernels automatically."
  },
  {
    "repo_name": "ome",
    "repo_link": "https://github.com/sgl-project/ome",
    "category": "container orchestration",
    "github_about_section": "OME is a Kubernetes operator for enterprise-grade management and serving of Large Language Models (LLMs)",
    "homepage_link": "http://docs.sglang.ai/ome"
  },
  {
    "repo_name": "OLMo-core",
    "repo_link": "https://github.com/allenai/OLMo-core",
    "category": "training framework",
    "github_about_section": "PyTorch building blocks for the OLMo ecosystem",
    "homepage_link": "https://olmo-core.readthedocs.io/en/latest/"
  },
  {
    "repo_name": "mistral-inference",
    "repo_link": "https://github.com/mistralai/mistral-inference",
    "category": "inference engine",
    "github_about_section": "Official inference library for Mistral models",
    "homepage_link": "https://mistral.ai"
  },
  {
    "repo_name": "triSYCL",
    "repo_link": "https://github.com/triSYCL/triSYCL",
    "category": "parallel computing",
    "github_about_section": "Generic system-wide modern C++ for heterogeneous platforms with SYCL from Khronos Group",
    "homepage_link": "https://trisycl.github.io/triSYCL/Doxygen/triSYCL/html/index.html"
  },
  {
    "repo_name": "tritonparse",
    "repo_link": "https://github.com/meta-pytorch/tritonparse",
    "category": "performance testing",
    "github_about_section": "TritonParse: A Compiler Tracer, Visualizer, and Reproducer for Triton Kernels",
    "homepage_link": "https://meta-pytorch.org/tritonparse"
  },
  {
    "repo_name": "StreamDiffusion",
    "repo_link": "https://github.com/cumulo-autumn/StreamDiffusion",
    "category": "image generation",
    "github_about_section": "StreamDiffusion: A Pipeline-Level Solution for Real-Time Interactive Generation",
    "homepage_link": "https://arxiv.org/abs/2312.12491"
  },
  {
    "repo_name": "reference-kernels",
    "repo_link": "https://github.com/gpu-mode/reference-kernels",
    "category": "kernel examples",
    "github_about_section": "Official Problem Sets / Reference Kernels for the GPU MODE Leaderboard!",
    "homepage_link": "https://gpumode.com"
  },
  {
    "repo_name": "hatchet",
    "repo_link": "https://github.com/LLNL/hatchet",
    "category": "performance testing",
    "github_about_section": "Graph-indexed Pandas DataFrames for analyzing hierarchical performance data",
    "homepage_link": "https://llnl-hatchet.readthedocs.io"
  },
  {
    "repo_name": "kernelbot",
    "repo_link": "https://github.com/gpu-mode/kernelbot",
    "category": "kernel examples",
    "github_about_section": "Write a fast kernel and see how you compare against the best humans and AI on gpumode.com",
    "homepage_link": "https://www.gpumode.com"
  },
  {
    "repo_name": "cutile-python",
    "repo_link": "https://github.com/NVIDIA/cutile-python",
    "category": "parallel computing",
    "github_about_section": "cuTile is a programming model for writing parallel kernels for NVIDIA GPUs",
    "homepage_link": "https://docs.nvidia.com/cuda/cutile-python"
  },
  {
    "repo_name": "FTorch",
    "repo_link": "https://github.com/Cambridge-ICCS/FTorch",
    "category": "middleware",
    "github_about_section": "A library for directly calling PyTorch ML models from Fortran.",
    "homepage_link": "https://cambridge-iccs.github.io/FTorch"
  },
  {
    "repo_name": "KernelBench",
    "repo_link": "https://github.com/ScalingIntelligence/KernelBench",
    "category": "benchmark",
    "github_about_section": "KernelBench: Can LLMs Write GPU Kernels? - Benchmark with Torch -> CUDA problems",
    "homepage_link": "https://scalingintelligence.stanford.edu/blogs/kernelbench"
  },
  {
    "repo_name": "nvshmem",
    "repo_link": "https://github.com/NVIDIA/nvshmem",
    "category": "distributed computing",
    "github_about_section": "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.",
    "homepage_link": "https://docs.nvidia.com/nvshmem/api/index.html"
  },
  {
    "repo_name": "flashinfer-bench",
    "repo_link": "https://github.com/flashinfer-ai/flashinfer-bench",
    "category": "benchmark",
    "github_about_section": "Building the Virtuous Cycle for AI-driven LLM Systems",
    "homepage_link": "https://bench.flashinfer.ai"
  },
  {
    "repo_name": "Primus-Turbo",
    "repo_link": "https://github.com/AMD-AGI/Primus-Turbo",
    "category": "training framework",
    "github_about_section": "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 computation (FP8), and compute–communication overlap kernels."
  },
  {
    "repo_name": "BitBLAS",
    "repo_link": "https://github.com/microsoft/BitBLAS",
    "category": "Basic Linear Algebra Subprograms (BLAS)",
    "github_about_section": "BitBLAS is a library to support mixed-precision matrix multiplications, especially for quantized LLM deployment."
  },
  {
    "repo_name": "Wan2.2",
    "repo_link": "https://github.com/Wan-Video/Wan2.2",
    "category": "video generation",
    "github_about_section": "Wan: Open and Advanced Large-Scale Video Generative Models",
    "homepage_link": "https://wan.video"
  },
  {
    "repo_name": "kernels-community",
    "repo_link": "https://github.com/huggingface/kernels-community",
    "category": "gpu kernels",
    "homepage_link": "https://huggingface.co/kernels-community",
    "github_about_section": "Kernel sources for https://huggingface.co/kernels-community"
  },
  {
    "repo_name": "omnitrace",
    "repo_link": "https://github.com/ROCm/omnitrace",
    "category": "performance testing",
    "github_about_section": "Omnitrace: Application Profiling, Tracing, and Analysis",
    "homepage_link": "https://rocm.docs.amd.com/projects/omnitrace"
  },
  {
    "repo_name": "synthetic-data-kit",
    "repo_link": "https://github.com/meta-llama/synthetic-data-kit",
    "category": "synthetic data generation",
    "github_about_section": "Tool for generating high quality Synthetic datasets",
    "homepage_link": "https://pypi.org/project/synthetic-data-kit"
  },
  {
    "repo_name": "cudnn-frontend",
    "repo_link": "https://github.com/NVIDIA/cudnn-frontend",
    "category": "parallel computing",
    "github_about_section": "cudnn_frontend provides a c++ wrapper for the cudnn backend API and samples on how to use it",
    "homepage_link": "https://developer.nvidia.com/cudnn"
  },
  {
    "repo_name": "PipelineRL",
    "repo_link": "https://github.com/ServiceNow/PipelineRL",
    "category": "reinforcement learning",
    "github_about_section": "A scalable asynchronous reinforcement learning implementation with in-flight weight updates.",
    "homepage_link": "https://arxiv.org/abs/2509.19128"
  },
  {
    "repo_name": "cosmos-predict2.5",
    "repo_link": "https://github.com/nvidia-cosmos/cosmos-predict2.5",
    "category": "world model",
    "github_about_section": "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.",
    "homepage_link": "https://research.nvidia.com/labs/cosmos-lab/cosmos-predict2.5"
  },
  {
    "repo_name": "kraken",
    "repo_link": "https://github.com/meta-pytorch/kraken",
    "category": "kernel examples",
    "github_about_section": "Triton-based Symmetric Memory operators and examples"
  },
  {
    "repo_name": "TileIR",
    "repo_link": "https://github.com/microsoft/TileIR",
    "category": "parallel computing dsl",
    "github_about_section": "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 productivity without sacrificing the low-level optimizations necessary for state-of-the-art performance."
  },
  {
    "repo_name": "intelliperf",
    "repo_link": "https://github.com/AMDResearch/intelliperf",
    "category": "performance testing",
    "github_about_section": "Automated bottleneck detection and solution orchestration",
    "homepage_link": "https://arxiv.org/html/2508.20258v1"
  },
  {
    "repo_name": "streamv2v",
    "repo_link": "https://github.com/Jeff-LiangF/streamv2v",
    "category": "video generation",
    "github_about_section": "Official Pytorch implementation of StreamV2V.",
    "homepage_link": "https://jeff-liangf.github.io/projects/streamv2v"
  },
  {
    "repo_name": "tilus",
    "repo_link": "https://github.com/NVIDIA/tilus",
    "category": "parallel computing",
    "github_about_section": "Tilus is a tile-level kernel programming language with explicit control over shared memory and registers.",
    "homepage_link": "https://nvidia.github.io/tilus"
  },
  {
    "repo_name": "gemlite",
    "repo_link": "https://github.com/dropbox/gemlite",
    "category": "gpu kernels",
    "github_about_section": "Fast low-bit matmul kernels in Triton"
  },
  {
    "repo_name": "Self-Forcing",
    "repo_link": "https://github.com/guandeh17/Self-Forcing",
    "category": "video generation",
    "github_about_section": "Official codebase for \"Self Forcing: Bridging Training and Inference in Autoregressive Video Diffusion\" (NeurIPS 2025 Spotlight)",
    "homepage_link": "https://self-forcing.github.io"
  },
  {
    "repo_name": "TritonBench",
    "repo_link": "https://github.com/thunlp/TritonBench",
    "category": "benchmark",
    "github_about_section": "TritonBench: Benchmarking Large Language Model Capabilities for Generating Triton Operators",
    "homepage_link": "https://arxiv.org/abs/2502.14752"
  },
  {
    "repo_name": "IMO2025",
    "repo_link": "https://github.com/harmonic-ai/IMO2025",
    "category": "formal mathematical reasoning",
    "github_about_section": "Harmonic's model Aristotle achieved gold medal performance, solving 5 problems. This repository contains the lean statement files and proofs for Problems 1-5.",
    "homepage_link": "https://harmonic.fun"
  },
  {
    "repo_name": "RaBitQ",
    "repo_link": "https://github.com/gaoj0017/RaBitQ",
    "category": "quantization",
    "github_about_section": "[SIGMOD 2024] RaBitQ: Quantizing High-Dimensional Vectors with a Theoretical Error Bound for Approximate Nearest Neighbor Search",
    "homepage_link": "https://github.com/VectorDB-NTU/RaBitQ-Library"
  },
  {
    "repo_name": "torchdendrite",
    "repo_link": "https://github.com/sandialabs/torchdendrite",
    "category": "machine learning framework",
    "github_about_section": "Dendrites for PyTorch and SNNTorch neural networks"
  },
  {
    "repo_name": "triton-runner",
    "repo_link": "https://github.com/toyaix/triton-runner",
    "category": "debugger",
    "github_about_section": "Multi-Level Triton Runner supporting Python, IR, PTX, and cubin.",
    "homepage_link": "https://triton-runner.org"
  }
]