59 lines
2.3 KiB
Bash
Executable File
59 lines
2.3 KiB
Bash
Executable File
#!/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
|
||
# 注意：gpu051 已经在跑 official suite 的 aime25/aime26/live_code_bench，不要中断它。
|
||
# 它跑完后继续跑 hle。
|
||
|
||
# Machine 1 (gpu048): SWE + Agent（约 30-50h，取决于 SWE 镜像）
|
||
M1_BENCHES="swe_bench_verified,tau2_bench"
|
||
M1_FOLDER="DP4-flash-int8-thinking-gpu048"
|
||
|
||
# Machine 2 (gpu049): 知识与长上下文（约 12-18h）
|
||
M2_BENCHES="mmlu_pro,gpqa_diamond,longbench_v2"
|
||
M2_FOLDER="DP4-flash-int8-thinking-gpu049"
|
||
|
||
# Machine 3 (gpu051): 数学/代码 multi-run + hle（约 30h；hle 在 multi-run 跑完后启动）
|
||
M3_BENCHES="aime25,aime26,live_code_bench,hle"
|
||
M3_FOLDER="DP4-flash-int8-thinking-gpu051"
|
||
|
||
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 "==================================================="
|