thinletter embedding-quantization
Copyright 2026 Jan Rosecký (thinletter)

This product is licensed under the Apache License, Version 2.0 (see LICENSE). It includes or derives from the
following third-party components:

1. @wllama/wllama 3.6.1 (client/browser/vendor/wllama/: index.js, wllama.wasm; libllama build b10663-83d855c)
   Copyright (c) 2024 Xuan Son NGUYEN. MIT License (client/browser/vendor/wllama/LICENCE).
   llama.cpp / ggml inside the wasm: Copyright (c) 2023-2026 The ggml authors. MIT License.

2. @huggingface/tokenizers 0.2.0 (client/vqweb/vendor/tokenizers/tokenizers.mjs)
   Copyright 2018- The Hugging Face team. Apache License, Version 2.0 (client/vqweb/vendor/tokenizers/LICENSE).

3. client/vqweb/embed.js contains a Q2_K block decoder ported operation by operation from llama.cpp
   gguf-py/gguf/quants.py (Q2_K.dequantize_blocks). Portions derived from llama.cpp,
   Copyright (c) 2023-2026 The ggml authors, MIT License:

   Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
   documentation files (the "Software"), to deal in the Software without restriction, including without limitation
   the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and
   to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above
   copyright notice and this permission notice shall be included in all copies or substantial portions of the
   Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT
   LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT
   SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
   OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
   DEALINGS IN THE SOFTWARE.

4. The scalar export packs GGUF files with gguf-py (llama.cpp, MIT) as an imported library; the native evaluation uses
   llama.cpp binaries built from third_party/ (not distributed here).

Methods: GPTQ (Frantar et al.), GPTVQ (van Baalen et al.), Hadamard/QuaRot-style rotations and the other borrowed ideas
are credited in research/prior_art.md. Derived model weights published
on Hugging Face keep the licence of their base model (microsoft/harrier-oss-v1-0.6b MIT, Qwen/Qwen3-Embedding-0.6B
Apache-2.0, BAAI/bge-m3 MIT) with attribution in each model card.
