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, 17:15 UTC
[CI/Build][AMD] Enable Entrypoints Integration Test (Pooling) to run without error on ROCm (#29212)
Signed-off-by: Randall Smith <ransmith@amd.com> Co-authored-by: Randall Smith <ransmith@amd.com>
rasmith committed
Nov 21, 2025 at 20:13 UTC
6f403501a085f4917e49e1714bdf44d2aabd06f9
11 files changed
+66
tests/entrypoints/pooling/correctness/test_mteb_embed.py
+6
@@ -11,6 +11,12 @@ from tests.models.language.pooling_mteb_test.mteb_utils import (
11
run_mteb_embed_task,
12
)
13
from tests.utils import RemoteOpenAIServer
14
+from vllm.platforms import current_platform
15
+
16
+if current_platform.is_rocm():
17
+ pytest.skip(
18
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
19
+ )
20
21
os.environ["VLLM_LOGGING_LEVEL"] = "WARNING"
22
tests/entrypoints/pooling/correctness/test_mteb_score.py
+6
@@ -13,6 +13,12 @@ from tests.models.language.pooling_mteb_test.mteb_utils import (
13
run_mteb_rerank,
14
)
15
from tests.utils import RemoteOpenAIServer
16
+from vllm.platforms import current_platform
17
+
18
+if current_platform.is_rocm():
19
+ pytest.skip(
20
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
21
+ )
22
23
os.environ["VLLM_LOGGING_LEVEL"] = "WARNING"
24
tests/entrypoints/pooling/llm/test_embedding.py
+6
@@ -9,6 +9,12 @@ import torch.nn.functional as F
9
10
from vllm import LLM, PoolingParams
11
from vllm.distributed import cleanup_dist_env_and_memory
12
+from vllm.platforms import current_platform
13
+
14
+if current_platform.is_rocm():
15
+ pytest.skip(
16
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
17
+ )
18
19
MODEL_NAME = "intfloat/multilingual-e5-small"
20
tests/entrypoints/pooling/llm/test_encode.py
+6
@@ -7,6 +7,12 @@ import pytest
7
8
from vllm import LLM, PoolingParams
9
from vllm.distributed import cleanup_dist_env_and_memory
10
+from vllm.platforms import current_platform
11
+
12
+if current_platform.is_rocm():
13
+ pytest.skip(
14
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
15
+ )
16
17
MODEL_NAME = "intfloat/multilingual-e5-small"
18
tests/entrypoints/pooling/llm/test_score.py
+6
@@ -9,6 +9,12 @@ import torch
9
from tests.models.utils import softmax
10
from vllm import LLM, PoolingParams
11
from vllm.distributed import cleanup_dist_env_and_memory
12
+from vllm.platforms import current_platform
13
+
14
+if current_platform.is_rocm():
15
+ pytest.skip(
16
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
17
+ )
18
19
MODEL_NAME = "tomaarsen/Qwen3-Reranker-0.6B-seq-cls"
20
tests/entrypoints/pooling/openai/test_embedding.py
+6
@@ -19,6 +19,7 @@ from vllm.entrypoints.openai.protocol import (
19
EmbeddingResponse,
20
PoolingResponse,
21
)
22
+from vllm.platforms import current_platform
23
from vllm.transformers_utils.tokenizer import get_tokenizer
24
from vllm.utils.serial_utils import (
25
EMBED_DTYPE_TO_TORCH_DTYPE,
@@ -28,6 +29,11 @@ from vllm.utils.serial_utils import (
29
decode_pooling_output,
30
)
31
32
+if current_platform.is_rocm():
33
+ pytest.skip(
34
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
35
+ )
36
+
37
MODEL_NAME = "intfloat/multilingual-e5-small"
38
DUMMY_CHAT_TEMPLATE = """{% for message in messages %}{{message['role'] + ': ' + message['content'] + '\\n'}}{% endfor %}""" # noqa: E501
39
DTYPE = "bfloat16"
tests/entrypoints/pooling/openai/test_embedding_dimensions.py
+6
@@ -12,6 +12,12 @@ from tests.models.language.pooling.embed_utils import run_embedding_correctness_
12
from tests.models.utils import EmbedModelInfo
13
from tests.utils import RemoteOpenAIServer
14
from vllm.entrypoints.openai.protocol import EmbeddingResponse
15
+from vllm.platforms import current_platform
16
+
17
+if current_platform.is_rocm():
18
+ pytest.skip(
19
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
20
+ )
21
22
MODELS = [
23
EmbedModelInfo("intfloat/multilingual-e5-small", is_matryoshka=False),
tests/entrypoints/pooling/openai/test_embedding_long_text.py
+6
@@ -16,6 +16,12 @@ import pytest_asyncio
16
17
from tests.utils import RemoteOpenAIServer
18
from vllm.entrypoints.openai.protocol import EmbeddingResponse
19
+from vllm.platforms import current_platform
20
+
21
+if current_platform.is_rocm():
22
+ pytest.skip(
23
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
24
+ )
25
26
27
def _generate_random_text(word_count: int) -> str:
tests/entrypoints/pooling/openai/test_rerank.py
+6
@@ -8,6 +8,12 @@ import torch.nn.functional as F
8
9
from tests.utils import RemoteOpenAIServer
10
from vllm.entrypoints.openai.protocol import PoolingResponse, RerankResponse
11
+from vllm.platforms import current_platform
12
+
13
+if current_platform.is_rocm():
14
+ pytest.skip(
15
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
16
+ )
17
18
MODEL_NAME = "BAAI/bge-reranker-base"
19
DTYPE = "bfloat16"
tests/entrypoints/pooling/openai/test_score.py
+6
@@ -10,6 +10,12 @@ from torch import tensor
10
11
from tests.utils import RemoteOpenAIServer
12
from vllm.entrypoints.openai.protocol import ScoreResponse
13
+from vllm.platforms import current_platform
14
+
15
+if current_platform.is_rocm():
16
+ pytest.skip(
17
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
18
+ )
19
20
MODELS = [
21
{"name": "BAAI/bge-reranker-v2-m3", "is_cross_encoder": True},
tests/entrypoints/pooling/openai/test_truncation.py
+6
@@ -7,6 +7,12 @@ import pytest
7
import pytest_asyncio
8
9
from tests.utils import RemoteOpenAIServer
10
+from vllm.platforms import current_platform
11
+
12
+if current_platform.is_rocm():
13
+ pytest.skip(
14
+ "Encoder self-attention is not implemented on ROCm.", allow_module_level=True
15
+ )
16
17
MODEL_NAME = "sentence-transformers/all-MiniLM-L12-v2"
18
max_model_len = 128