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 16, 23:15 UTC
[Misc] Improve conftest (#5681)
Cyrus Leung committed
Jun 20, 2024 at 10:09 UTC
3730a1c832bca5ca8128aec3c7659304895edf2e
1 file changed
+3
-1
tests/conftest.py
+3
-1
@@ -365,7 +365,7 @@ class HfRunner:
365
cleanup()
366
367
368
-@pytest.fixture
368
+@pytest.fixture(scope="session")
369
def hf_runner():
370
return HfRunner
371
@@ -385,6 +385,7 @@ class VllmRunner:
385
block_size: int = 16,
386
enable_chunked_prefill: bool = False,
387
swap_space: int = 4,
388
+ enforce_eager: bool = False,
389
**kwargs,
390
) -> None:
391
self.model = LLM(
@@ -393,6 +394,7 @@ class VllmRunner:
394
trust_remote_code=True,
395
dtype=dtype,
396
swap_space=swap_space,
397
+ enforce_eager=enforce_eager,
398
disable_log_stats=disable_log_stats,
399
tensor_parallel_size=tensor_parallel_size,
400
max_model_len=max_model_len,