From cb67ac86217fd4a555c8fedd8a99ccfcb912fc5d Mon Sep 17 00:00:00 2001 From: Gahow Wang Date: Thu, 23 Jul 2026 16:24:20 +0800 Subject: [PATCH] Reuse validated FlashInfer cache for TP2 smoke --- runs/frontier-tp2-prefill-serving-v0/experiment-card.md | 6 ++++++ runs/frontier-tp2-prefill-serving-v0/jobs-smoke.toml | 8 ++++---- 2 files changed, 10 insertions(+), 4 deletions(-) diff --git a/runs/frontier-tp2-prefill-serving-v0/experiment-card.md b/runs/frontier-tp2-prefill-serving-v0/experiment-card.md index c3bd6f8..843cfa7 100644 --- a/runs/frontier-tp2-prefill-serving-v0/experiment-card.md +++ b/runs/frontier-tp2-prefill-serving-v0/experiment-card.md @@ -71,6 +71,12 @@ - **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` 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`,约 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 次 profiler-off serving + warmup,所有计时发生在 warmup 后。 ## 结果 diff --git a/runs/frontier-tp2-prefill-serving-v0/jobs-smoke.toml b/runs/frontier-tp2-prefill-serving-v0/jobs-smoke.toml index b00891f..63a8e22 100644 --- a/runs/frontier-tp2-prefill-serving-v0/jobs-smoke.toml +++ b/runs/frontier-tp2-prefill-serving-v0/jobs-smoke.toml @@ -1,16 +1,16 @@ version = 1 [[jobs]] -name = "tp2-prefill-serving-smoke-20260723-v0" +name = "tp2-prefill-serving-smoke-20260723-v1" gpus = 2 gpu_model = "H20" hosts = ["dash1"] 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] TP = "2" 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" -FLASHINFER_WORKSPACE_BASE = "/tmp/frontier-tp2-prefill-serving-v0-flashinfer" +FLASHINFER_WORKSPACE_BASE = "/tmp/frontier-component-flashinfer-v4"