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 14, 05:30 UTC
[Bug] Fix Test in Batch Invariant (#26128)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Oct 8, 2025 at 13:13 UTC
4ba88757497b250e78d67e318a661b69e7b46da3
1 file changed
+4
-1
tests/v1/generation/test_batch_invariance.py
+4
-1
@@ -292,8 +292,11 @@ def LLM_with_max_seqs(
292
# Allow some CPU offload if needed.
293
swap_space=swap_space,
294
# Keep things lean and CI-friendly.
295
- dtype="float16",
295
+ dtype="auto",
296
# Single-GPU by default; override externally if desired.
297
tensor_parallel_size=int(os.getenv("VLLM_TP_SIZE", "1")),
298
trust_remote_code=os.getenv("VLLM_TRUST_REMOTE_CODE", "0") == "1",
299
+ enable_prefix_caching=False,
300
+ # Enable for MOE models
301
+ # enable_expert_parallel=True,
302
)