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 6, 00:45 UTC
[ROCm][CI] Fix flaky Cohere/OpenAI embedding parity test (#37616)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Mar 25, 2026 at 05:55 UTC
f262a62aa191b4fc3070a5fd7f418246cb326f06
4 files changed
+8
-4
tests/entrypoints/pooling/embed/test_cohere_openai_parity.py
+2
-2
@@ -10,7 +10,7 @@ import numpy as np
10
import pytest
11
import requests
12
13
-from tests.utils import RemoteOpenAIServer
13
+from tests.utils import ROCM_EXTRA_ARGS, RemoteOpenAIServer
14
15
MODEL_NAME = "BAAI/bge-base-en-v1.5"
16
DTYPE = "bfloat16"
@@ -28,7 +28,7 @@ def server():
28
"512",
29
"--gpu-memory-utilization",
30
"0.02",
31
- ]
31
+ ] + ROCM_EXTRA_ARGS
32
with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:
33
yield remote_server
34
tests/entrypoints/pooling/embed/test_online_dimensions.py
+2
-2
@@ -10,7 +10,7 @@ import pytest
10
from tests.conftest import HfRunner
11
from tests.models.language.pooling.embed_utils import run_embedding_correctness_test
12
from tests.models.utils import EmbedModelInfo
13
-from tests.utils import RemoteOpenAIServer
13
+from tests.utils import ROCM_EXTRA_ARGS, RemoteOpenAIServer
14
from vllm.entrypoints.pooling.embed.protocol import EmbeddingResponse
15
from vllm.platforms import current_platform
16
@@ -49,7 +49,7 @@ def server(model_info, dtype: str):
49
"--enforce-eager",
50
"--max-model-len",
51
"512",
52
- ]
52
+ ] + ROCM_EXTRA_ARGS
53
54
if model_info.name == "Snowflake/snowflake-arctic-embed-m-v1.5":
55
# Manually enable Matryoshka Embeddings
vllm/entrypoints/pooling/base/serving.py
+1
@@ -118,6 +118,7 @@ class PoolingServing:
118
)
119
120
pooling_params = self.io_processor.create_pooling_params(ctx.request)
121
+ pooling_params.verify(self.model_config)
122
123
for i, engine_prompt in enumerate(ctx.engine_prompts):
124
prompt_request_id = (
vllm/entrypoints/utils.py
+3
@@ -309,6 +309,9 @@ def create_error_response(
309
310
if isinstance(message, Exception):
311
exc = message
312
+ logger.debug(
313
+ "create_error_response called with %s: %s", type(exc).__name__, exc
314
+ )
315
316
from vllm.exceptions import VLLMNotFoundError, VLLMValidationError
317