COMPLIANCE · LICENSES

Third-party licenses

Third-party component acknowledgments for MiniLLM. MiniLLM itself is freeware and not open source; it downloads or invokes these components at runtime. Model weights are never bundled.

Effective · 2026-09-23Scope · MiniLLM v1.1.0Contact · bashigestudio@mail.ru

CONTENTS 1. How MiniLLM uses third-party software 2. llama.cpp & runtime binaries 3. Build variants 4. Model files 5. Trademarks 6. Obtaining full license texts

How MiniLLM uses third-party software

MiniLLM is a free GUI manager; it may download prebuilt llama.cpp releases and launch llama-server with user-selected parameters.
The app itself ships as an executable only (no source). Third-party OSS remains under its original licenses.
No LLM weights are included in MiniLLM.exe — see Terms §5.

llama.cpp & runtime binaries

Component	Role	License (typical)
llama.cpp (ggml-org)	Inference engine, `llama-server`, OpenAI-compatible API	MIT
ggml	Tensor library used by llama.cpp	MIT
Vulkan / CUDA / SYCL / ROCm / OpenVINO builds	GPU/accelerator backends of llama.cpp	See upstream
Windows / system TLS roots	HTTPS downloads; MiniLLM may refresh root certs on new PCs	OS trust store

Upstream: github.com/ggml-org/llama.cpp — license files ship with each released build. Verify the exact license bundled with the build you install (llama-server / Install tab shows the build id).

Build variants

MiniLLM can install any of the following llama.cpp packages; backend libraries inside each package remain under their upstream licenses:

cpu-x64 — CPU only (Intel/AMD)
vulkan — Vulkan GPU path
cuda — NVIDIA CUDA (also uses CUDA runtime libraries under NVIDIA license terms)
sycl — Intel oneAPI SYCL
rocm — AMD ROCm
openvino — Intel OpenVINO
arm64 — Windows on ARM

Model files

MiniLLM does not bundle any model. GGUF language models and mmproj vision projectors you download are licensed by their publishers (model card / repo license — e.g. Apache-2.0, MIT, Llama license, Qwen license…).

You must comply with each model's copyright, acceptable-use policy, and export/distribution rules.
Quantized derivatives (UD-Q, Q4_K_XL, …) usually inherit the base model's license — confirm on the model page.
Responsibility for model licensing rests with the user — see Terms §5.

Trademarks

Names and logos of NVIDIA, AMD, Intel, Microsoft, Vulkan, ROCm, CUDA, SYCL, OpenVINO, OpenAI, Qwen, LLaVA and others are trademarks of their respective owners. MiniLLM is the product name of independent software by Bashige (not a registered trademark of Bashige) and is not endorsed by these companies. “OpenAI-compatible” describes API shape only, not affiliation.

Obtaining full license texts

Installed llama.cpp: license files are placed alongside the binaries by the installer (Install tab path).
Upstream source licenses: repository of each project on GitHub.
Models: license file in the model repo / GGUF release notes.
Requests: bashigestudio@mail.ru

Related pages: Terms · Disclaimer · Privacy