Back to Gap Map Infrastructure / Compilers & Model Optimization

MLC LLM

open source / Overall score: 3.5

MLC LLM compiles large language models for native execution across a wide range of consumer and server hardware, then runs them on its own MLCEngine. Compilation targets Vulkan, CUDA, ROCm, Metal, OpenCL, WebGPU and WASM, covering Linux, Windows, macOS, iOS, Android and the browser. The project builds on the MLC fork of Apache TVM's compilation stack.

Openness

5 high confidence

5.0

license: Apache-2.0(OSI)
source: public
core features withheld: no

The LICENSE file contains the Apache-2.0 text. The repository is public and the README describes no paid tier, enterprise edition or license-gated build, so the core is ungated.

https://raw.githubusercontent.com/mlc-ai/mlc-llm/main/CMakeLists.txt recorded 2026-09-30
Top-level CMakeLists.txt builds the engine from repository sources with "tvm_file_glob(GLOB_RECURSE MLC_LLM_SRCS cpp/*.cc)" and links the TVM runtime from "set(TVM_SOURCE_DIR 3rdparty/tvm)". Searched the body for enterprise, license key, paid, commercial and proprietary: none.
https://raw.githubusercontent.com/mlc-ai/mlc-llm/main/LICENSE recorded 2026-09-30
LICENSE file is the verbatim Apache-2.0 text, opening with "Apache License" and "Version 2.0, January 2004" and "TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION".
https://raw.githubusercontent.com/mlc-ai/mlc-llm/main/README.md recorded 2026-09-30
README carries a license badge "license-apache_2" and describes "MLC LLM is a machine learning compiler and high-performance deployment engine for large language models." Searched the body for enterprise, license key, paid, commercial, proprietary and non-commercial: none.

Adoption

3 low confidence

3.0

MLC LLM distributes prebuilt wheels from its own package index rather than PyPI, which publishes no download figure. That leaves GitHub stars as the only adoption signal, measuring attention rather than confirmed use.

https://api.github.com/repos/mlc-ai/mlc-llm recorded 2026-08-18
Repo metadata - stargazers_count = 23,071 - for mlc-ai/mlc-llm.

Capability

4 medium confidence

4.0

MLC LLM compiles LLMs through the MLC fork of the Apache TVM stack to Vulkan, CUDA, ROCm, Metal, OpenCL, WebGPU and WASM across six platforms, then serves them from its own engine. Target breadth is wide, but the model class is narrow, one step below Apache TVM's fully general scope.

https://raw.githubusercontent.com/mlc-ai/mlc-llm/main/.gitmodules recorded 2026-09-30
The .gitmodules file vendors the TVM stack as a submodule, with "path = 3rdparty/tvm" pointing at "url = https://github.com/mlc-ai/relax.git", a fork rather than apache/tvm.
https://raw.githubusercontent.com/mlc-ai/mlc-llm/main/CMakeLists.txt recorded 2026-09-30
Top-level CMakeLists.txt builds the engine from repository sources with "tvm_file_glob(GLOB_RECURSE MLC_LLM_SRCS cpp/*.cc)" and links the TVM runtime from "set(TVM_SOURCE_DIR 3rdparty/tvm)". Searched the body for enterprise, license key, paid, commercial and proprietary: none.
https://raw.githubusercontent.com/mlc-ai/mlc-llm/main/README.md recorded 2026-09-30
README shows a platform matrix with "Vulkan, ROCm", "Vulkan, CUDA", "Metal (dGPU)", "WebGPU and WASM", "Metal on Apple A-series GPU" and "OpenCL on Adreno GPU", and says "MLCEngine provides OpenAI-compatible API available through REST server, python, javascript, iOS, Android".

Verified 2026-09-27