flashinfer-python has 4.5M monthly downloads and 6K GitHub stars. Version 0.6.14 supports Python <4.0,>=3.10, first published in 2025.
It is actively maintained under a permissive license, last released in Jul 2026.
It is built on top of apache-tvm-ffi, click and cuda-tile. It is used by 19 tracked packages, including sglang, vllm and tensorrt-llm.
GitHub topics include gpu, large-large-models and cuda.
Package Insights
((week_daily_avg - month_daily_avg) / month_daily_avg) * 100Weekly Downloads
GitHub Stars
Downloads by OS
Python Versions
Top Countries
Dependencies
- apache-tvm-ffi !=0.1.8,!=0.1.8.post0,<0.2,>=0.1.6
- click
- cuda-tile >=1.4.0
- einops
- ninja
- numpy
- nvidia-cudnn-frontend >=1.13.0
- nvidia-cutlass-dsl >=4.5.0
- nvidia-ml-py
- packaging >=24.2
- requests
- tabulate
- torch
- tqdm
3 optional dependencies
- cuda-python[nvep]
- nvidia-cutlass-dsl[cu13]
- nvidia-cutlass-dsl[cu12]