sora/evalstone
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update multi-machine distribution: move hle to gpu051

Browse Source
This commit is contained in:
sora 2026-07-28 02:40:37 +00:00
parent 5ab3f249c6
commit ee27130d4d
1 changed files with 12 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
scripts/run_multi_machine.sh
View File

@ -21,17 +21,20 @@ COMMON_ARGS="--model DeepSeek-V4-Flash-Int8 --api-url http://localhost:30000/v1
# 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"
# 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 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 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 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"
# 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"