[CI] Temporarily disable GB300 jobs (#36981)

.github/workflows/_pr-test-stage.yml

@@ -100,8 +100,10 @@ jobs:
   run:
     name: ${{ inputs.self_name }} (${{ matrix.partition }})
     # Runs on schedule / parallel-dispatch / non-failed PR with main_package or sgl_kernel changes.
+    # Temporarily skip the broken GB300 runner before GitHub tries to allocate it.
     if: |
       always() &&
+      inputs.runner_config != '4-gpu-gb300' &&
       ((github.event_name == 'schedule' || fromJson(inputs.caller_inputs).test_parallel_dispatch == true) || (!failure() && !cancelled())) &&
       (fromJson(inputs.check_changes).main_package == 'true' || fromJson(inputs.check_changes).sgl_kernel == 'true')
     # runs-on resolved from runs_on_map; check-changes already substituted

.github/workflows/release-whl-deepgemm.yml

@@ -194,9 +194,10 @@ jobs:
           # - arch_label: sm120
           #   runner: 1-gpu-5090
           #   wheel_arch: x86_64
-          - arch_label: sm100-aarch64
+          # Temporarily disabled while the GB300 runner is unavailable.
-            runner: 4-gpu-gb300
+          # - arch_label: sm100-aarch64
-            wheel_arch: aarch64
+          #   runner: 4-gpu-gb300
+          #   wheel_arch: aarch64
     runs-on: ${{ matrix.runner }}
     timeout-minutes: 120
     steps:

.github/workflows/rerun-test.yml

@@ -87,7 +87,8 @@ permissions:
 
 jobs:
   rerun-test-cuda:
-    if: inputs.mode == 'cuda'
+    # Temporarily reject reruns targeting the broken GB300 runner.
+    if: inputs.mode == 'cuda' && inputs.runs_on != '4-gpu-gb300'
     runs-on: ${{ inputs.runs_on }}
     timeout-minutes: 120
     permissions:
@@ -199,7 +200,8 @@ jobs:
         if: failure()
 
   rerun-test-multimodal-gen:
-    if: inputs.mode == 'multimodal_gen'
+    # Temporarily reject reruns targeting the broken GB300 runner.
+    if: inputs.mode == 'multimodal_gen' && inputs.runs_on != '4-gpu-gb300'
     runs-on: ${{ inputs.runs_on }}
     timeout-minutes: 120
     permissions: