vllm.cpp
Copyright 2026 Ettore Di Giacinto
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
This product is a C++ port of vLLM (https://github.com/vllm-project/vllm),
which is licensed under the Apache License, Version 2.0. It incorporates,
ports, and/or vendors code from third-party projects, including:
- vLLM (Apache-2.0) — the upstream project this is a 1:1 port of.
- Marlin / GPTQ-Marlin mixed-precision kernels (Apache-2.0), as vendored
by vLLM under src/vt/cuda/marlin/.
- NVIDIA CUTLASS (BSD-3-Clause) — used for FP4/FP8 GEMM kernels.
- FlashAttention-2 (BSD-3-Clause) — vendored CUDA attention sources under
src/vt/cuda/flash_attn/.
- Flash Linear Attention (MIT) — vendored Triton kernel sources under
triton_kernels/.
- xgrammar (Apache-2.0) — C++ grammar core, vendored for structured output.
- cpp-httplib (MIT) and nlohmann/json (MIT) — header-only dependencies.
- google/minja (MIT) - header-only Jinja engine vendored under
third_party/minja/ (with its LICENSE), used as the chat-template renderer
in src/vllm/entrypoints/chat_template.cpp.
Each such component remains under its own license; the notices and license
texts of vendored third-party code are retained in their respective source
files and directories.
TRADEMARKS
vllm.cpp is an independent community project. It is NOT affiliated with,
endorsed by, or sponsored by the vLLM project, the PyTorch Foundation, or the
Linux Foundation.
"vLLM" is a mark of its respective owners and is used in this repository in its
descriptive sense only, to identify the upstream project that vllm.cpp ports,
mirrors, and is gated against. No claim of ownership of that mark is made, and
nothing here implies that the vLLM project has reviewed, approved, or endorsed
this software. Section 6 of the Apache License, Version 2.0 grants no trademark
rights, and none are claimed.
Other names and marks referenced here (llama.cpp, SGLang, MLX, CUTLASS,
FlashInfer, Marlin, TRT-LLM, and others) belong to their respective owners and
are likewise used descriptively.