# Third-party notices

This project is an independent implementation and does not link third-party inference runtimes.
The following MIT-licensed source was studied and used as an algorithm/layout reference:

- **llama.cpp / ggml**, Copyright (c) Georgi Gerganov and contributors, MIT License.
  Relevant references include the GGUF v2/v3 reader, Gemma 4 model graph and converter, raw-UTF-8
  BPE tokenizer, FP16 conversion behavior, Q4_0/Q8_0 block layouts, and Metal RMS-reduction
  organization. Upstream source:
  <https://github.com/ggml-org/llama.cpp>

The full upstream MIT license is available at
<https://github.com/ggml-org/llama.cpp/blob/master/LICENSE>.