| [ |
| { |
| "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" |
| } |
| ] |
