biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CPU] Support for Whisper (#30062)

Browse Source 
Signed-off-by: Aditya Tewari <aditya.tewari@arm.com>
This commit is contained in:
Aditya Tewari
2025-12-10 12:58:42 +00:00
committed by GitHub
parent 53d2420b44
commit cebda2a4af
5 changed files with 49 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
tests/models/multimodal/generation/test_whisper.py
View File

@@ -92,13 +92,14 @@ def run_test(
*,
tensor_parallel_size: int,
distributed_executor_backend: str | None = None,
dtype: str = "half",
) -> None:
prompt_list = PROMPTS * 10
expected_list = EXPECTED[model] * 10
with vllm_runner(
model,
dtype="half",
dtype=dtype,
max_model_len=448,
tensor_parallel_size=tensor_parallel_size,
distributed_executor_backend=distributed_executor_backend,
@@ -120,12 +121,28 @@ def run_test(
@pytest.mark.core_model
@pytest.mark.parametrize("model", ["openai/whisper-large-v3-turbo"])
@pytest.mark.parametrize("dtype", ["half"])
@create_new_process_for_each_test()
def test_models(vllm_runner, model) -> None:
def test_models(vllm_runner, model, dtype) -> None:
run_test(
vllm_runner,
model,
tensor_parallel_size=1,
dtype=dtype,
)
@pytest.mark.cpu_model
@pytest.mark.parametrize("model", ["openai/whisper-large-v3-turbo"])
@pytest.mark.parametrize("dtype", ["half"])
def test_models_cpu(vllm_runner, model, dtype) -> None:
# @create_new_process_for_each_test() does not work for some runners
# TODO: to fix cpu privilege issues in run-cpu-test-arm.sh
run_test(
vllm_runner,
model,
tensor_parallel_size=1,
dtype=dtype,
)