「Intel MacでBonsai-8Bを動かす」を編集中


== 環境 ==
* Mac Pro 2013 OCLP Sequoia
* 古すぎてMetalが使えないのでOpenBLASとGGMLを使う。

== 手順 ==
=== homebrew で必要なものを入れる ===
<source lang=bash>
% brew install cmake pkg-config openblas uv
</source>

=== llama.cpp bonsai カスタム版を入れる ===

<source lang=bash>
% git clone https://github.com/PrismML-Eng/llama.cpp

% cmake -B build \
  -DGGML_BLAS=ON \
  -DGGML_BLAS_VENDOR=OpenBLAS \
  -DCMAKE_PREFIX_PATH="/usr/local/opt/openblas" \
  -DBLAS_LIBRARIES="/usr/local/opt/openblas/lib/libopenblas.dylib" \
  -DBLAS_INCLUDE_DIR="/usr/local/opt/openblas/include"

% cmake --build build --config Release --parallel $(sysctl -n hw.ncpu)

cd ..
</source>

=== bonsai のモデルをダウンロード ===
<source lang=bash>
% brew install uv
% uvx hf download prism-ml/Bonsai-8B-gguf --local-dir ./models
</source>

## cli で動かす
<source lang=bash>
% ./llama.cpp/build/bin/llama-cli -m models/Bonsai-8B.gguf
</source>

## Web UI / Web API を動かす
<source lang=bash>
% ./llama.cpp/build/bin/llama-server \
    -m ./models/Bonsai-8B.gguf \
    -ngl 99 \
    --host 0.0.0.0 \
    --port 18080 \
    --api-key local-bonsai \
    --alias bonsai-8b
</source>

Web APIが動いているか確認する
 http://localhost:18080/v1/models

こんな感じの応答があれば動いている
<source lang="json">
{"models":[{"name":"bonsai-8b","model":"bonsai-8b","modified_at":"","size":"","digest":"","type":"model","description":"","tags":[""],"capabilities":["completion"],"parameters":"","details":{"parent_model":"","format":"gguf","family":"","families":[""],"parameter_size":"","quantization_level":""}}],"object":"list","data":[{"id":"bonsai-8b","aliases":["bonsai-8b"],"tags":[],"object":"model","created":1775728104,"owned_by":"llamacpp","meta":{"vocab_type":2,"n_vocab":151669,"n_ctx_train":65536,"n_embd":4096,"n_params":8188548096,"size":1152704128}}]}
</source>

== 結論 ==
これだけ古いMacを動かすのは、電気代の無駄であり、まったくエコではない。

なにも得られるものはない。

[[category: 人工知能]]
[[category: Mac]]