Intel Arc A770で Strata を動かそう

# ------------------------------------------------------------------------------
# 1. oneAPI 環境変数の読み込み & リポジトリ取得
# ------------------------------------------------------------------------------
source /opt/intel/oneapi/setvars.sh

cd ~/install/Strata_test
git clone https://github.com/Niko1221/Strata.git
cd Strata

# ------------------------------------------------------------------------------
# 2. uv による仮想環境作成 & 依存パッケージ導入
# ------------------------------------------------------------------------------
uv venv
source .venv/bin/activate
uv pip install -r requirements.txt

# setup_intel.py 内部の "python -m pip" 呼び出しに対応するため pip を追加
uv pip install pip

# ------------------------------------------------------------------------------
# 3. SYCL バックエンドのネイティブビルド
# ------------------------------------------------------------------------------
chmod +x sycl/tools/build.sh
./sycl/tools/build.sh

# ------------------------------------------------------------------------------
# 4. モデル設定とデータ準備
# ------------------------------------------------------------------------------
python sycl/setup_intel.py
# [対話プロンプトでの入力内容]
# - Which model?: 1 (Qwen3.8-Flash-Next)
# - Which size?: 4 (IQ3_S)
# - Context?: 3 (65536 tokens)
# - KV cache?: 1 (8-bit)
# - Turn on the experimental speed projection?: n (off)

starting run-iq3_s.sh ...

loading the model (the first start takes a minute or two) ...

[strata] starting the engine: reading the model's weights ...

Traceback (most recent call last):

  File "/home/test/install/Strata/sycl/serve/server_intel.py", line 294, in <module>

    sys.exit(S.main())

             ^^^^^^^^

  File "/home/test/install/Strata/serve/server.py", line 5436, in main

    engine = StrataEngine(exe, engine_args(cfg) + (effort_end or []), cwd=cfg.get("cwd"), log=cfg.get("log"),

             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "/home/test/install/Strata/serve/server.py", line 676, in __init__

    raise RuntimeError(why + (f" (see {log})" if log else "") +

RuntimeError: the engine exited before it was ready (see /home/test/install/Strata/strata-iq3_s.log)

the engine log's last lines:

  [strata] 2026-10-08 04:54:47 engine started: /home/test/install/Strata/sycl/serve/strata-sycl.sh --serve --pack /work/Strata-data/packs/iq3_s --native /work/Strata-data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf --ple-gguf /work/Strata-data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf --expert-profile /work/Strata/data/expert-profile.bin --expert-cache auto --spec 4 --spec-min-p 0.5 --mtp /work/Strata-data/mtp/rt --max-context 65536 --kv int8 --pool-workers 11 --kv-resident 32768 --ple-io ram --vram-reserve-mib 300 --prefill 4096

  /home/test/install/Strata/sycl/serve/strata-sycl.sh: 37 行: exec: docker: 見つかりません

# ------------------------------------------------------------------------------
# 5. 設定ファイルをホスト（ネイティブ実行）向けに書き換え
# ------------------------------------------------------------------------------
# バックアップの作成
cp strata-iq3_s.json strata-iq3_s.json.bak

# Docker ラッパーをホストのビルド済みバイナリに差し替え、/work/ を実パスに置換
sed -i 's|sycl/serve/strata-sycl.sh|build-sycl/strata|g' strata-iq3_s.json
sed -i 's|/work/|/home/test/install/|g' strata-iq3_s.json

# ------------------------------------------------------------------------------
# 6. サーバーの起動
# ------------------------------------------------------------------------------
./run-iq3_s.sh

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp

git fetch origin refs/pull/28243/head:pr-28243-mtp
git checkout pr-28243-mtp