Add deployment and multi-machine benchmark scripts

- scripts/build_and_upload_docker.sh: sync context, build, save and upload
  evalscope-complete-py312 image to ModelScope.
- scripts/deploy_remote_machine.sh: SSH to a fresh machine and run install.sh.
- scripts/run_multi_machine.sh: distribute official-suite benchmarks across
  3 machines (gpu048, gpu049, gpu051).

scripts/build_and_upload_docker.sh

@@ -0,0 +1,66 @@
+#!/bin/bash
+# ============================================================
+# 构建并上传 evalscope-complete-py312 Docker 镜像
+# 用法: bash scripts/build_and_upload_docker.sh
+# ============================================================
+
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+DOCKER_CTX="$ROOT_DIR/tools/docker"
+IMAGE_TAG="evalscope-complete-py312:latest"
+OUTPUT_TAR="$ROOT_DIR/docker/evalscope-complete-py312.tar.gz"
+LOG_FILE="$ROOT_DIR/logs/build_docker_$(date +%Y%m%d_%H%M%S).log"
+
+mkdir -p "$ROOT_DIR/logs"
+mkdir -p "$ROOT_DIR/docker"
+
+exec > >(tee -a "$LOG_FILE")
+exec 2>&1
+
+echo "==> 开始构建 Docker 镜像: $IMAGE_TAG"
+echo "    日志: $LOG_FILE"
+echo ""
+
+# 1. 同步构建上下文
+echo "==> 1. 同步构建上下文到 $DOCKER_CTX"
+cd "$ROOT_DIR"
+rsync -av --delete --exclude='*.log' --exclude='__pycache__' --exclude='.git' \
+    bash/ "$DOCKER_CTX/bash/"
+rsync -av --delete --exclude='*.log' --exclude='__pycache__' --exclude='.git' \
+    evalscope/ "$DOCKER_CTX/evalscope/"
+rsync -av --delete --exclude='*.log' --exclude='__pycache__' --exclude='.git' \
+    tools/tau2-bench/ "$DOCKER_CTX/tau2-bench/"
+
+# 2. 构建镜像
+echo ""
+echo "==> 2. 构建 Docker 镜像"
+cd "$DOCKER_CTX"
+docker build -f Dockerfile.py312 -t "$IMAGE_TAG" .
+
+# 3. 保存镜像
+echo ""
+echo "==> 3. 保存镜像到 $OUTPUT_TAR"
+docker save "$IMAGE_TAG" | gzip > "$OUTPUT_TAR"
+ls -lh "$OUTPUT_TAR"
+
+# 4. 生成 md5
+echo ""
+echo "==> 4. 生成 md5"
+md5sum "$OUTPUT_TAR" > "$OUTPUT_TAR.md5"
+
+# 5. 上传到 ModelScope
+echo ""
+echo "==> 5. 上传到 ModelScope"
+modelscope upload \
+    --commit-message "Update evalscope-complete-py312 with torch (CPU) for tokenizer support" \
+    SoraAmami/evalscope-docker \
+    "$OUTPUT_TAR"
+
+echo ""
+echo "========================================"
+echo "构建并上传完成！"
+echo "镜像: $IMAGE_TAG"
+echo "本地 tar: $OUTPUT_TAR"
+echo "日志: $LOG_FILE"
+echo "========================================"

scripts/deploy_remote_machine.sh

@@ -0,0 +1,29 @@
+#!/bin/bash
+# ============================================================
+# 在远程机器上部署 EvalScope
+# 用法: bash scripts/deploy_remote_machine.sh <IP> [BASE_DIR]
+# 示例: bash scripts/deploy_remote_machine.sh 174.1.51.3 /data1/sora
+# ============================================================
+
+set -euo pipefail
+
+IP="${1:?请提供远程机器 IP}"
+BASE_DIR="${2:-/data1/sora}"
+PASS="${SSHPASS:-sskj2025}"
+REMOTE="root@$IP"
+
+SSH="sshpass -p $PASS ssh -o StrictHostKeyChecking=no -o ConnectTimeout=10 $REMOTE"
+SCP="sshpass -p $PASS scp -o StrictHostKeyChecking=no -o ConnectTimeout=10"
+
+echo "==> 部署到 $IP ..."
+
+# 1. 确保基础目录存在
+$SSH "mkdir -p $BASE_DIR && which git docker || (apt-get update && apt-get install -y git docker.io)"
+
+# 2. 克隆/更新代码
+$SSH "cd $BASE_DIR && if [ -d evalscope ]; then cd evalscope && git pull; else git clone https://git.meta-stone.net/sora/evalstone.git evalscope; fi"
+
+# 3. 运行部署脚本（下载数据、Docker 镜像、配置环境）
+$SSH "cd $BASE_DIR/evalscope && bash install.sh $BASE_DIR"
+
+echo "==> $IP 部署完成"

scripts/run_multi_machine.sh

@@ -0,0 +1,55 @@
+#!/bin/bash
+# ============================================================
+# 多机器 benchmark 分发脚本
+# 根据当前进度把 official suite 拆到 3 台机器跑
+# 用法: bash scripts/run_multi_machine.sh
+# ============================================================
+
+set -euo pipefail
+
+PASS="${SSHPASS:-sskj2025}"
+SSH="sshpass -p $PASS ssh -o StrictHostKeyChecking=no -o ConnectTimeout=10"
+
+# 机器配置
+MACHINES=(
+    "174.1.51.1:/data1/sora"   # gpu048 / P800-01
+    "174.1.51.2:/data1/sora"   # gpu049 / P800-02
+    "174.1.51.4:/data1/sora"   # gpu051 / P800-04
+)
+
+COMMON_ARGS="--model DeepSeek-V4-Flash-Int8 --api-url http://localhost:30000/v1 --dataset-dir /data1/sora/evalscope --output-dir /data1/sora/evalscope/output --batch-size 4 --thinking --limit none --seed 42"
+
+# 分发方案：尽量让 3 台机器同时跑完
+# 估算：aime25/aime26 12 run，live_code_bench 5 run，其余 1 run
+# Machine 1: 推理/数学 multi-run（约 30h）
+M1_BENCHES="aime25,aime26,live_code_bench"
+M1_FOLDER="DP4-flash-int8-thinking-m1"
+
+# Machine 2: 知识与长上下文（约 18h）
+M2_BENCHES="hle,mmlu_pro,gpqa_diamond,longbench_v2"
+M2_FOLDER="DP4-flash-int8-thinking-m2"
+
+# Machine 3: SWE + Agent（约 30-50h，取决于 SWE 镜像）
+M3_BENCHES="swe_bench_verified,tau2_bench"
+M3_FOLDER="DP4-flash-int8-thinking-m3"
+
+run_on_machine() {
+    local ip_base="$1"
+    local ip="${ip_base%%:*}"
+    local base_dir="${ip_base##*:}"
+    local datasets="$2"
+    local folder="$3"
+    local cmd="cd $base_dir/evalscope && nohup python bash/run.py --datasets $datasets --folder-name $folder $COMMON_ARGS > $base_dir/evalscope/logs/run_${folder}.log 2>&1 &"
+
+    echo "==> 在 $ip 启动: $folder ($datasets)"
+    $SSH root@$ip "$cmd"
+}
+
+run_on_machine "${MACHINES[0]}" "$M1_BENCHES" "$M1_FOLDER"
+run_on_machine "${MACHINES[1]}" "$M2_BENCHES" "$M2_FOLDER"
+run_on_machine "${MACHINES[2]}" "$M3_BENCHES" "$M3_FOLDER"
+
+echo ""
+echo "==================================================="
+echo "已分发到 3 台机器，各自日志在 logs/run_*.log"
+echo "==================================================="