A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 12, 22:15 UTC
[ROCm][CI] Fix Weight Loading With Multiple GPU Tests on ROCm (#28984)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Micah Williamson committed
Nov 19, 2025 at 15:31 UTC
22e44ad589d951f440ef98141a2a6f9df97f6873
3 files changed
+8
-3
.buildkite/test-amd.yaml
+2
-3
@@ -1323,7 +1323,7 @@ steps:
1323
- vllm/
1324
- tests/weight_loading
1325
commands:
1326
- - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models.txt
1326
+ - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-amd.txt
1327
1328
- label: Weight Loading Multiple GPU Test - Large Models # optional
1329
mirror_hardwares: [amdexperimental]
@@ -1331,13 +1331,12 @@ steps:
1331
# grade: Blocking
1332
working_dir: "/vllm-workspace/tests"
1333
num_gpus: 2
1334
- gpu: a100
1334
optional: true
1335
source_file_dependencies:
1336
- vllm/
1337
- tests/weight_loading
1338
commands:
1340
- - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large.txt
1339
+ - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large-amd.txt
1340
1341
- label: NixlConnector PD accuracy tests (Distributed) # 30min
1342
mirror_hardwares: [amdexperimental]
tests/weight_loading/models-amd.txt
new
+3
@@ -0,0 +1,3 @@
1
+fp8, amd/Meta-Llama-3.1-8B-Instruct-FP8-KV, main
2
+None, amd/Llama-3.2-1B-Instruct-FP8-KV, main
3
+fp8, amd/Mixtral-8x7B-Instruct-v0.1-FP8-KV, main
tests/weight_loading/models-large-amd.txt
new
+3
@@ -0,0 +1,3 @@
1
+fp8, amd/Meta-Llama-3.1-70B-Instruct-FP8-KV, main
2
+None, microsoft/phi-4, main
3
+fp8, amd/Mixtral-8x22B-Instruct-v0.1-FP8-KV, main