[Klaud Cold] Update qwen3.5-bf16-b200-sglang (+mtp) SGLang image to v0.5.12-cu130 (#1446)

* Update qwen3.5-bf16-b200-sglang (+mtp +agentic) SGLang image to v0.5.12-cu130

The recipe was pinned to lmsysorg/sglang:nightly-dev-20260216-d3bae71e —
86 days old. Bumps all three variants of the qwen3.5-bf16-b200-sglang
recipe family to the stable v0.5.12-cu130 release tag (already used by
other b200 sglang recipes on main).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore: fill in pr-link for #1446 in perf-changelog

* Drop qwen3.5-bf16-b200-sglang-agentic from this PR

Comment out the agentic-coding sibling recipe and remove it from the
changelog entry. The agentic flow is blocked by an e2e-tests.yml /
benchmark-tmpl.yml artifact-name mismatch (downloads agentic_* but
uploads as bmk_agentic_*) and shouldn't be exercised here. PR now only
bumps the non-agentic and -mtp variants.

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: functionstackx

.github/configs/nvidia-master.yaml

@@ -2052,7 +2052,7 @@ dsv4-fp4-b300-sglang-mtp:
       - { tp: 4, ep: 1, conc-start: 4, conc-end: 32, spec-decoding: mtp }
 
 qwen3.5-bf16-b200-sglang:
-  image: lmsysorg/sglang:nightly-dev-20260216-d3bae71e
+  image: lmsysorg/sglang:v0.5.12-cu130
   model: Qwen/Qwen3.5-397B-A17B
   model-prefix: qwen3.5
   runner: b200
@@ -2071,7 +2071,7 @@ qwen3.5-bf16-b200-sglang:
       - { tp: 8, ep: 1, conc-start: 4, conc-end: 64 }
 
 qwen3.5-bf16-b200-sglang-mtp:
-  image: lmsysorg/sglang:nightly-dev-20260216-d3bae71e
+  image: lmsysorg/sglang:v0.5.12-cu130
   model: Qwen/Qwen3.5-397B-A17B
   model-prefix: qwen3.5
   runner: b200
@@ -2089,24 +2089,22 @@ qwen3.5-bf16-b200-sglang-mtp:
       search-space:
       - { tp: 8, ep: 1, conc-start: 4, conc-end: 64, spec-decoding: mtp }
 
-# Diverged from qwen3.5-bf16-b200-sglang (agentic-coding sibling). Metadata is
-# identical to origin/main's qwen3.5-bf16-b200-sglang; the split exists because this
-# PR adds an agentic-coding scenarios block that differs from main
-# (either main had none or had a different conc/offload sweep).
-# The original qwen3.5-bf16-b200-sglang entry stays byte-identical to origin/main.
-qwen3.5-bf16-b200-sglang-agentic:
-  image: lmsysorg/sglang:nightly-dev-20260216-d3bae71e
-  model: Qwen/Qwen3.5-397B-A17B
-  model-prefix: qwen3.5
-  runner: b200
-  precision: bf16
-  framework: sglang
-  multinode: false
-  scenarios:
-    agentic-coding:
-    - duration: 1800
-      search-space:
-      - { tp: 8, ep: 1, offloading: none, conc-list: [1, 2, 4, 8, 16, 32] }
+# agentic-coding sibling — temporarily disabled, blocked by e2e-tests.yml
+# artifact-name mismatch (downloads `agentic_*` but benchmark-tmpl.yml uploads
+# as `bmk_agentic_*`). Re-enable once that workflow is aligned.
+# qwen3.5-bf16-b200-sglang-agentic:
+#   image: lmsysorg/sglang:v0.5.12-cu130
+#   model: Qwen/Qwen3.5-397B-A17B
+#   model-prefix: qwen3.5
+#   runner: b200
+#   precision: bf16
+#   framework: sglang
+#   multinode: false
+#   scenarios:
+#     agentic-coding:
+#     - duration: 1800
+#       search-space:
+#       - { tp: 8, ep: 1, offloading: none, conc-list: [1, 2, 4, 8, 16, 32] }
 
 qwen3.5-fp8-b200-sglang:
   image: lmsysorg/sglang:nightly-dev-20260422-de962f32

perf-changelog.yaml

@@ -2667,3 +2667,10 @@
     - "Update SGLang image from v0.5.11-cu130 to v0.5.12-cu130"
     - "Temporarily disable agentic-coding scenario (blocked by e2e-tests.yml artifact-name mismatch)"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1415
+
+- config-keys:
+    - qwen3.5-bf16-b200-sglang
+    - qwen3.5-bf16-b200-sglang-mtp
+  description:
+    - "Update SGLang image from nightly-dev-20260216-d3bae71e (86d old) to v0.5.12-cu130"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1446