[CI Failure] fix test_default_mm_loras (vllm-project#27795)

Signed-off-by: Huamin Li <[email protected]>

Author: hl475

tests/lora/test_default_mm_loras.py

@@ -30,7 +30,8 @@
     "enable_lora": "True",
     "max_num_seqs": 2,
     "max_lora_rank": 320,
-    "max_model_len": 12800,
+    # Keep these LoRA tests on short-RoPE for determinism post-LongRoPE change.
+    "max_model_len": 4096,
     "gpu_memory_utilization": 0.8,
     "limit_mm_per_prompt": {"audio": 1},
     "enforce_eager": True,