[ROCm][CI] Fix failure in Language Models Tests (Extra Standard) by reducing...

[ROCm][CI] Fix failure in Language Models Tests (Extra Standard) by reducing agent pool size (#31553)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>

.buildkite/test-amd.yaml

@@ -859,7 +859,7 @@ steps:
 - label: Language Models Tests (Extra Standard) %N
   timeout_in_minutes: 45
   mirror_hardwares: [amdexperimental]
-  agent_pool: mi325_8
+  agent_pool: mi325_2
   # grade: Blocking
   torch_nightly: true
   source_file_dependencies:
@@ -871,6 +871,7 @@ steps:
     # Shard slow subset of standard language models tests. Only run when model
     # source is modified, or when specified test files are modified
     - pip freeze | grep -E 'torch'
+    - export TORCH_NCCL_BLOCKING_WAIT=1
     - pytest -v -s models/language -m 'core_model and slow_test' \
              --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT \
              --shard-id=$$BUILDKITE_PARALLEL_JOB