Fix Gemma eval native context a7ed9610

EVAL-QUEUE-RUNBOOK.md

@@ -107,3 +107,13 @@ six-row mechanics JSON in the fresh bundle workspace, then run the copied
 eval worker will wait while mechanics owns memory. GPU0 can evaluate meanwhile.
 The tokenizer regression uses CPU mocks; the corrected real-model canary has
 not yet run on the company node.
+
+## Gemma 2 context correction
+
+The launcher now uses the native 8192-token context. Before GPU startup, the
+existing runtime tokenizer checks every exact chat prompt plus its full output
+cap. Overflow fails explicitly: no hidden truncation, omitted questions or
+long-context override. The earlier 16384 launcher setting was incorrect.
+`recover-startup --from-plan OLD --out NEW` supports only failed startup queues
+with no results, reuses verified prepared data, and preserves the original
+start/admission/deadline timestamps. It does not renew the 20-hour budget.

simct-b200-portable.bundle

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10f42cb4b625116aeb5a492eee42364a181527171d90895e19ee941fb58b11ed
-size 4159222
+oid sha256:6aafb042e138f63cf5d558a9d9bda737665330f194b9f3e894c12b45949faa2b
+size 4163162

source-manifest.json

@@ -1,9 +1,9 @@
 {
   "source_branch": "vdt/ops/b200-portable",
-  "source_commit": "bbfef64f0d65e117b97595e704316e4afe642b8e",
+  "source_commit": "a7ed9610c2de02bab040d30136d2bfe8f574f8e7",
   "bundle_file": "simct-b200-portable.bundle",
-  "bundle_sha256": "10f42cb4b625116aeb5a492eee42364a181527171d90895e19ee941fb58b11ed",
-  "bundle_bytes": 4159222,
+  "bundle_sha256": "6aafb042e138f63cf5d558a9d9bda737665330f194b9f3e894c12b45949faa2b",
+  "bundle_bytes": 4163162,
   "history": "complete",
   "format": "git-bundle",
   "reason": "HF Git receive hook rejects historical PNG/JPG blobs; bundle preserves original Git object IDs",
@@ -20,6 +20,6 @@
   },
   "eval_queue": {
     "pull-eval-queue.sh": "d20d18a6b22bae31313f13b26252f7032a290e4b229af842ddfe3a0d77ad8aa8",
-    "EVAL-QUEUE-RUNBOOK.md": "d66a8a79cd202626dea2998a6b28069eb9c9018a01794f8938aee14bb9c39b02"
+    "EVAL-QUEUE-RUNBOOK.md": "e249ed6750fb4264496fd6ec7bf2a5dcc67ccfc3be4e7f85be05cb00d8446f2b"
   }
 }