# apache/tvm

Open Machine Learning Compiler Framework

Repository: https://github.com/apache/tvm
Canonical: https://ross.abutalabs.com/products/tvm
Homepage: https://tvm.apache.org/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: compiler, tensor, deep-learning, gpu, opencl, metal, performance, javascript, rocm, tvm, vulkan, spirv, machine-learning
Last push: 2026-08-26T13:47:21+00:00

## Health v2 (maintenance only)
Score: 90/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 73, longevity 100
- inputs: {"age_days": 3612, "days_push": 7, "days_rel": 23, "gap_med": 92.0, "n_releases_24m": 9}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 13691, forks 3959 (observed 2026-08-28T04:11:04.588753+00:00)

## What it is
Apache TVM is an open machine learning compiler framework that takes pre-trained models and compiles them into optimized, deployable modules with minimal runtimes. It follows a Python-first development model and supports universal deployment from data center GPUs to edge devices.

## Use cases
- compile a trained PyTorch model for deployment on edge devices
- optimize deep learning model inference for a specific GPU
- generate minimal deployable runtime modules from ML models
- auto-tune tensor kernels for custom hardware backends
- deploy LLM inference with a compiled minimal runtime
- cross-compile models for embedded or mobile targets
- customize ML compiler pipelines in Python

## When to choose
- you need to deploy ML models across diverse hardware (GPU, CPU, mobile, embedded)
- you want minimal runtime overhead and small deployable artifacts
- you need to optimize or auto-tune kernels for custom or emerging hardware
- you want a Python-first compiler API to customize compilation pipelines

## When to avoid
- you just need a simple inference runtime without compilation or tuning
- your target is only mainstream hardware well-served by vendor libraries like cuDNN or TensorRT
- you need a plug-and-play solution with minimal compiler expertise on the team

## Facets
- artifact type: framework
- maturity: active
- function: compiler, machine-learning, llm-inference, gpu-computing, benchmarking
- domain: machine-learning, compilers, deep-learning, gpu-computing, performance
- platform: python, cpp, windows, cross-platform, wasm
- tags: ml-compilation, tensor-compiler, code-generation, model-deployment, auto-tuning, tensorir, relax, edge-deployment, opencl, vulkan, metal, rocm, spirv, linux, macos, gpu, android, ios

## Member repositories
- apache/tvm (main) score 90

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:11:04.588753+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:13:06.849068+00:00, confidence not recorded.
  - readme: https://github.com/apache/tvm (fetched 2026-08-28T04:11:04.588753+00:00, sha b381df9dc9b1)
  - homepage: https://tvm.apache.org/ (fetched 2026-08-29T08:07:32.384972+00:00, sha e11ddc8ddd43)
  - site_page: https://tvm.apache.org/docs/reference/security.html (fetched 2026-08-29T08:07:32.395953+00:00, sha dce930925d1a)
  - site_page: https://tvm.apache.org/docs (fetched 2026-08-29T08:07:32.393984+00:00, sha 6740040a44a1)
- Data as of 2026-08-30T08:39:29.467469+00:00.