[Core] Whisper enable FULL_DECODE_ONLY CudaGraph (#30072)
Signed-off-by: NickLucche <nlucches@redhat.com>
This commit is contained in:
@@ -103,6 +103,8 @@ def run_test(
|
||||
max_model_len=448,
|
||||
tensor_parallel_size=tensor_parallel_size,
|
||||
distributed_executor_backend=distributed_executor_backend,
|
||||
# TODO (NickLucche) figure out output differences with non-eager and re-enable
|
||||
enforce_eager=True,
|
||||
) as vllm_model:
|
||||
llm = vllm_model.llm
|
||||
|
||||
|
||||
Reference in New Issue
Block a user