Reuse validated FlashInfer cache for TP2 smoke

This commit is contained in:
2026-07-23 16:24:20 +08:00
parent fd859bc52c
commit cb67ac8621
2 changed files with 10 additions and 4 deletions

View File

@@ -71,6 +71,12 @@
- **Fleet sync deviation** dash1 未安装 `rsync`fleet mandatory sync 使用 - **Fleet sync deviation** dash1 未安装 `rsync`fleet mandatory sync 使用
`scp` 一个已校验为空的 `/tmp/frontier-tp2-prefill-serving-v0-empty-sync` `scp` 一个已校验为空的 `/tmp/frontier-tp2-prefill-serving-v0-empty-sync`
目录实验代码仍只来自上述 Git worktree 目录实验代码仍只来自上述 Git worktree
- **FlashInfer cache deviation** v0 使用空 workspace 冷构建
`fused_moe_90` 8 分钟仅完成 4/181 objects且尚未发请求因此主动
终止exit 141)。v1 改用此前 component profile 在同一 dash1同一
FlashInfer `0.6.8.post1/sm90a` 构建的
`/tmp/frontier-component-flashinfer-v4`仍保留 2 profiler-off serving
warmup所有计时发生在 warmup
## 结果 ## 结果

View File

@@ -1,16 +1,16 @@
version = 1 version = 1
[[jobs]] [[jobs]]
name = "tp2-prefill-serving-smoke-20260723-v0" name = "tp2-prefill-serving-smoke-20260723-v1"
gpus = 2 gpus = 2
gpu_model = "H20" gpu_model = "H20"
hosts = ["dash1"] hosts = ["dash1"]
command = "timeout --signal=TERM --kill-after=60s 3600 bash runs/frontier-tp2-prefill-serving-v0/run_prefill_profile.sh" command = "timeout --signal=TERM --kill-after=60s 3600 bash runs/frontier-tp2-prefill-serving-v0/run_prefill_profile.sh"
artifacts = ["runs/frontier-tp2-prefill-serving-v0/remote-outputs/tp2-smoke"] artifacts = ["runs/frontier-tp2-prefill-serving-v0/remote-outputs/tp2-smoke-r1"]
[jobs.env] [jobs.env]
TP = "2" TP = "2"
SERVER_PORT = "9827" SERVER_PORT = "9827"
OUTPUT_ROOT = "runs/frontier-tp2-prefill-serving-v0/remote-outputs/tp2-smoke" OUTPUT_ROOT = "runs/frontier-tp2-prefill-serving-v0/remote-outputs/tp2-smoke-r1"
WARMUP_REQUESTS = "2" WARMUP_REQUESTS = "2"
FLASHINFER_WORKSPACE_BASE = "/tmp/frontier-tp2-prefill-serving-v0-flashinfer" FLASHINFER_WORKSPACE_BASE = "/tmp/frontier-component-flashinfer-v4"