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, 00:30 UTC
[CI Failure] fix test_default_mm_loras (#27795)
Signed-off-by: Huamin Li <3ericli@gmail.com>
Huamin Li committed
Oct 30, 2025 at 03:13 UTC
c7d2a554baf8694503e6865b5df300650b6c6b6b
1 file changed
+2
-1
tests/lora/test_default_mm_loras.py
+2
-1
@@ -30,7 +30,8 @@ VLLM_RUNNER_BASE_KWARGS = {
30
"enable_lora": "True",
31
"max_num_seqs": 2,
32
"max_lora_rank": 320,
33
- "max_model_len": 12800,
33
+ # Keep these LoRA tests on short-RoPE for determinism post-LongRoPE change.
34
+ "max_model_len": 4096,
35
"gpu_memory_utilization": 0.8,
36
"limit_mm_per_prompt": {"audio": 1},
37
"enforce_eager": True,