96 lines
2.7 KiB
Bash
96 lines
2.7 KiB
Bash
#!/bin/bash
|
|
# DGX Spark Claude Code ローカル実行スクリプト
|
|
# Usage: curl -sL https://docs.techswan.online/scripts/dgx-spark-claude-code.sh | bash
|
|
|
|
set -e
|
|
|
|
echo "🚀 DGX Spark Claude Code セットアップ"
|
|
echo ""
|
|
|
|
# Step 1: vLLM起動
|
|
echo "📦 Step 1: vLLM + Qwen3-Coder-Next-FP8 起動..."
|
|
docker stop qwen3-coder-next-fp8 2>/dev/null || true
|
|
docker rm qwen3-coder-next-fp8 2>/dev/null || true
|
|
|
|
docker run -d --name qwen3-coder-next-fp8 \
|
|
--gpus all \
|
|
-p 8000:8000 \
|
|
-v ~/.cache/huggingface:/root/.cache/huggingface \
|
|
--ipc=host \
|
|
nvcr.io/nvidia/vllm:25.11-py3 \
|
|
vllm serve Qwen/Qwen3-Coder-Next-FP8 \
|
|
--served-model-name qwen3-coder-next \
|
|
--max-model-len 200000 \
|
|
--max-num-seqs 8 \
|
|
--gpu-memory-utilization 0.85 \
|
|
--enable-auto-tool-choice \
|
|
--tool-call-parser qwen3_xml \
|
|
--trust-remote-code
|
|
|
|
echo " コンテナ起動完了（モデルロードに約10分かかります）"
|
|
echo ""
|
|
|
|
# Step 2: claude-code-proxy セットアップ
|
|
echo "🔧 Step 2: claude-code-proxy セットアップ..."
|
|
PROXY_DIR="$HOME/.claude-code-proxy"
|
|
|
|
if [ ! -d "$PROXY_DIR" ]; then
|
|
git clone https://github.com/fuergaosi233/claude-code-proxy.git "$PROXY_DIR"
|
|
fi
|
|
|
|
cd "$PROXY_DIR"
|
|
pip install -r requirements.txt --user --quiet 2>/dev/null || pip install -r requirements.txt --user --break-system-packages --quiet
|
|
|
|
cat > .env << 'EOF'
|
|
OPENAI_BASE_URL=http://localhost:8000/v1
|
|
OPENAI_API_KEY=dummy
|
|
BIG_MODEL=qwen3-coder-next
|
|
MIDDLE_MODEL=qwen3-coder-next
|
|
SMALL_MODEL=qwen3-coder-next
|
|
REQUEST_TIMEOUT=300
|
|
MAX_TOKENS_LIMIT=8192
|
|
EOF
|
|
|
|
echo " プロキシ設定完了"
|
|
echo ""
|
|
|
|
# Step 3: プロキシ起動
|
|
echo "🌐 Step 3: プロキシ起動..."
|
|
pkill -f start_proxy.py 2>/dev/null || true
|
|
nohup python3 start_proxy.py > /tmp/proxy.log 2>&1 &
|
|
sleep 3
|
|
|
|
if curl -s http://localhost:8082/health > /dev/null 2>&1; then
|
|
echo " プロキシ起動完了 (port 8082)"
|
|
else
|
|
echo " ⚠️ プロキシ起動中... (数秒待ってください)"
|
|
fi
|
|
echo ""
|
|
|
|
# Step 4: vLLMヘルスチェック待ち
|
|
echo "⏳ Step 4: vLLM起動待ち..."
|
|
echo " （モデルロード中... 約10分かかります）"
|
|
echo ""
|
|
|
|
while ! curl -s http://localhost:8000/health > /dev/null 2>&1; do
|
|
sleep 10
|
|
echo -n "."
|
|
done
|
|
echo ""
|
|
echo " ✅ vLLM起動完了！"
|
|
echo ""
|
|
|
|
# 完了メッセージ
|
|
echo "=========================================="
|
|
echo "🎉 セットアップ完了！"
|
|
echo ""
|
|
echo "Claude Codeを起動するには:"
|
|
echo ""
|
|
echo " ANTHROPIC_BASE_URL=http://localhost:8082 ANTHROPIC_API_KEY=dummy claude"
|
|
echo ""
|
|
echo "または ~/.bashrc に追加:"
|
|
echo ""
|
|
echo " alias claude-local='ANTHROPIC_BASE_URL=http://localhost:8082 ANTHROPIC_API_KEY=dummy claude'"
|
|
echo ""
|
|
echo "=========================================="
|