[Bugfix][CI/Build][Hardware][AMD] Shard ID parameters in AMD tests ru…

…nning parallel jobs (vllm-project#9279) Signed-off-by: Hissu Hyvarinen <[email protected]> Signed-off-by: Sumit Dubey <[email protected]>
sumitd2 · Nov 14, 2024 · 4e6fe2f · 4e6fe2f
1 parent dc46c9a
commit 4e6fe2f
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 6 deletions.
diff --git a/.buildkite/run-amd-test.sh b/.buildkite/run-amd-test.sh
@@ -107,11 +107,12 @@ fi
 PARALLEL_JOB_COUNT=8
 # check if the command contains shard flag, we will run all shards in parallel because the host have 8 GPUs. 
 if [[ $commands == *"--shard-id="* ]]; then
+  # assign job count as the number of shards used   
+  commands=${commands//"--num-shards= "/"--num-shards=${PARALLEL_JOB_COUNT} "}
   for GPU in $(seq 0 $(($PARALLEL_JOB_COUNT-1))); do
-    #replace shard arguments
-    commands=${commands//"--shard-id= "/"--shard-id=${GPU} "}
-    commands=${commands//"--num-shards= "/"--num-shards=${PARALLEL_JOB_COUNT} "}
-    echo "Shard ${GPU} commands:$commands"
+    # assign shard-id for each shard
+    commands_gpu=${commands//"--shard-id= "/"--shard-id=${GPU} "}
+    echo "Shard ${GPU} commands:$commands_gpu"
     docker run \
         --device /dev/kfd --device /dev/dri \
         --network host \
@@ -123,7 +124,7 @@ if [[ $commands == *"--shard-id="* ]]; then
         -e HF_HOME=${HF_MOUNT} \
         --name ${container_name}_${GPU}  \
         ${image_name} \
-        /bin/bash -c "${commands}" \
+        /bin/bash -c "${commands_gpu}" \
         |& while read -r line; do echo ">>Shard $GPU: $line"; done &
     PIDS+=($!)
   done

diff --git a/tests/lora/test_minicpmv.py b/tests/lora/test_minicpmv.py
@@ -1,8 +1,11 @@
 from typing import List
 
+import pytest
+
 import vllm
 from vllm.assets.image import ImageAsset
 from vllm.lora.request import LoRARequest
+from vllm.platforms import current_platform
 
 MODEL_PATH = "openbmb/MiniCPM-Llama3-V-2_5"
 
@@ -53,6 +56,9 @@ def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int) -> List[str]:
     return generated_texts
 
 
+@pytest.mark.xfail(
+    current_platform.is_rocm(),
+    reason="MiniCPM-V dependency xformers incompatible with ROCm")
 def test_minicpmv_lora(minicpmv_lora_files):
     llm = vllm.LLM(
         MODEL_PATH,
@@ -63,7 +69,6 @@ def test_minicpmv_lora(minicpmv_lora_files):
         trust_remote_code=True,
         gpu_memory_utilization=0.97  # This model is pretty big for CI gpus
     )
-
     output1 = do_sample(llm, minicpmv_lora_files, lora_id=1)
     for i in range(len(EXPECTED_OUTPUT)):
         assert EXPECTED_OUTPUT[i].startswith(output1[i])