feat: PR#11 - EQ-Bench3 macOS評価セットアップ (uv + llama-cpp + HF Inference Providers)
#11
by YUGOROU - opened
概要
Vast.aiからMacBook M3への実行環境移行。uv仮想環境で完全隔離。
変更
eqbench-mac/setup_eqbench_mac.sh追加
構成
- 受験者: TeenEmo GGUF (llama-server, Metal加速, port 8000)
- 採点者: openai/gpt-oss-120b via HF Inference Providers / novita ($0.05/$0.25 per 1M tokens)
- 環境:
~/eqbench-teenemo/.venv(uv、iOSプロジェクトと完全隔離)
推定コスト
46シナリオ × ~8,000 input + ~500 output tokens ≈ $0.024 (Pro $2クレジット内)
YUGOROU changed pull request status to open
YUGOROU changed pull request status to merged