瀏覽代碼

shift notes: qwen pinned-image verdict — bench 94, live agent clean, dgx3 off

Alexander Korolev 3 周之前
父節點
當前提交
c31f0f107f
共有 1 個文件被更改,包括 7 次插入 和 0 次删除
  1. 7 0
      history/night-shifts/2026-09-26.md

+ 7 - 0
history/night-shifts/2026-09-26.md

@@ -285,3 +285,10 @@ bench baseline params restored (MTP k=4, atomic-add 1). Launcher fix:
 `get_image_digest` now accepts digest refs (`docker image ls` silently
 returns nothing for `@sha256:` refs — would have made every pinned prebuilt
 recipe fail). Re-bench pending (label int4-auto-1x-pinned).
+
+Verdict: re-bench on the pinned image scored **94/100 (164/174)** — same
+quality as the suspect-image baseline (93), so the fix costs nothing. The
+user's live agent session on dgx5.lan has since run clean ("works like a
+clock") — the long-context degeneration that had hit ~5 sessions/day is
+gone. dgx3.lan turned off after the bench; fleet: dgx5 (qwen3.8, pinned) +
+dgx1/2 (EXL3 2x) live.