[ROCm][CI] Fix Weight Loading With Multiple GPU Tests on ROCm (#28984)

Signed-off-by: Micah Williamson <micah.williamson@amd.com>

Author: micah-wil

.buildkite/test-amd.yaml

@@ -1323,21 +1323,20 @@ steps:
   - vllm/
   - tests/weight_loading
   commands:
-    - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models.txt
+    - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-amd.txt
 
 - label: Weight Loading Multiple GPU Test - Large Models # optional
   mirror_hardwares: [amdexperimental]
   agent_pool: mi325_2
   # grade: Blocking
   working_dir: "/vllm-workspace/tests"
   num_gpus: 2
-  gpu: a100
   optional: true
   source_file_dependencies:
   - vllm/
   - tests/weight_loading
   commands:
-    - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large.txt
+    - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large-amd.txt
 
 - label: NixlConnector PD accuracy tests (Distributed) # 30min
   mirror_hardwares: [amdexperimental]

tests/weight_loading/models-amd.txt

@@ -0,0 +1,3 @@
+fp8, amd/Meta-Llama-3.1-8B-Instruct-FP8-KV, main
+None, amd/Llama-3.2-1B-Instruct-FP8-KV, main
+fp8, amd/Mixtral-8x7B-Instruct-v0.1-FP8-KV, main

tests/weight_loading/models-large-amd.txt

@@ -0,0 +1,3 @@
+fp8, amd/Meta-Llama-3.1-70B-Instruct-FP8-KV, main
+None, microsoft/phi-4, main
+fp8, amd/Mixtral-8x22B-Instruct-v0.1-FP8-KV, main