IntermediatePython

FlashInferKernel Library for LLM Inference on NVIDIA GPUs

FlashInfer is an open-source CUDA library focused on LLM inference, providing attention, GEMM, and MoE kernels. It supports NVIDIA GPUs from Turing through Blackwell, is primarily written in Python, and is licensed under Apache-2.0. The project has gained 5876 stars on GitHub, indicating strong community interest.

5.9K Stars
1.1K Forks
734 Issues
85 Views
Python
Apache-2.0
Indexed

Project Overview

FlashInfer is an open-source CUDA library focused on LLM inference, providing attention, GEMM, and MoE kernels. It supports NVIDIA GPUs from Turing through Blackwell, is primarily written in Python, and is licensed under Apache-2.0. The project has gained 5876 stars on GitHub, indicating strong community interest.

Project Overview

FlashInfer is an open-source CUDA library designed to accelerate large language model (LLM) inference on NVIDIA GPUs. It includes various core kernels such as attention, GEMM, and MoE, aiming to improve inference performance.

Core Features

  • Optimized attention kernels for LLM inference.
  • Efficient GEMM (General Matrix Multiplication) kernels.
  • Kernel implementations for Mixture of Experts (MoE) models.

Technology Stack

The project is primarily written in Python, built on CUDA technology, and supports NVIDIA GPU architectures from Turing through Blackwell.

License

FlashInfer is licensed under Apache-2.0, allowing users freedom to use, modify, and distribute.

Open Source Status

As an open-source project, FlashInfer is publicly available on GitHub and currently has 5876 stars, reflecting an active community.

FlashInferLLM inference accelerationFlashAttentionPageAttentionCUDA kernelslarge model deploymentvLLMhigh-performance computingopen-source kernel libraryattention optimization

Project Rating

0.0 (0 Reviews)

Share

Frequently Asked Questions

What is FlashInfer: Kernel Library for LLM Inference on NVIDIA GPUs?

FlashInfer is an open-source CUDA library focused on LLM inference, providing attention, GEMM, and MoE kernels. It supports NVIDIA GPUs from Turing through Blackwell, is primarily written in Python, and is licensed under Apache-2.0. The project has gained 5876 stars on GitHub, indicating strong community interest.

What language is FlashInfer: Kernel Library for LLM Inference on NVIDIA GPUs written in?

FlashInfer: Kernel Library for LLM Inference on NVIDIA GPUs is primarily written in Python.

What license is FlashInfer: Kernel Library for LLM Inference on NVIDIA GPUs under?

FlashInfer: Kernel Library for LLM Inference on NVIDIA GPUs is released under the Apache-2.0 license.

Related Projects

No results yet

Explore More

Similar Tools

Cursor

Cursor

A smart code editor based on secondary development of VS Code, with "native built-in AI" as its core selling point. It does not rely on plugins but deeply integrates AI into the underlying architecture of the editor, enabling it to understand the context of the entire project's codebase. It also supports seamless migration of all VS Code configurations and plugins.

Google Antigravity

Google Antigravity

Antigravity supports multiple models, including Gemini 3 Pro, Claude Sonnet 4.5, and GPT-OSS, allowing developers to select the most suitable model for their tasks within the same environment.

Codex

Codex

OpenAI Codex is an AI programming model and assistant developed by OpenAI, capable of translating natural language instructions into corresponding source code. It provides developers with intelligent code completion and code generation functionalities. Initially launched in 2021 as the code model for the OpenAI API, it once served as the core engine for GitHub Copilot. With the evolution of OpenAI's technology, Codex returned in 2025 in a new form as an "AI programming agent," capable of understanding complex requirements and automatically writing and debugging code, significantly enhancing development efficiency and software delivery speed.

Kiro

Kiro

Kiro is an AI-powered programming IDE launched by AWS, which adopts a specification-driven development model. It transforms natural language requirements into clear specification documents and tasks, then uses built-in AI agents to generate code, debug, and optimize, providing comprehensive assistance throughout the development process of large-scale projects.

Trae

Trae

Trae (official website: trae.ai) is an AI-native integrated development environment (IDE) launched by ByteDance. It is not merely a programming assistant but rather a "collaborative partner" that deeply integrates large language models (LLMs) to help developers achieve more intelligent and automated software development—from requirements analysis and code construction to debugging and deployment.

Claude

Claude

Claude is an intelligent language interaction platform developed by the American AI company Anthropic. It integrates capabilities such as deep text understanding, information organization, code assistance, and task analysis, enabling it to handle more complex tasks beyond simple chat conversations. These include long-text summarization, image analysis, logical reasoning, and programming assistance, among others. Compared to some single-purpose Q&A bots, Claude functions more like an intelligent tool equipped with reasoning logic and scalable features.

Comments

Comments

0
0/500 Characters

No comments yet

Be the first to comment

Open Source Projects

Explore, learn and contribute to open source AI projects to advance the development of artificial intelligence technology

View All