Merge branch 'dev' of github.com:modelscope/twinkle into dev

Author: hjh0119

src/twinkle/sampler/vllm_sampler/vllm_engine.py

@@ -47,12 +47,12 @@ def __init__(
         model_id: str,
         *,
         tensor_parallel_size: int = 1,
-        gpu_memory_utilization: float = 0.9,
+        gpu_memory_utilization: float = 0.7,
         max_model_len: Optional[int] = None,
         max_num_seqs: int = 256,
         enable_lora: bool = True,
-        max_loras: int = 64,
-        max_lora_rank: int = 64,
+        max_loras: int = 5,
+        max_lora_rank: int = 32,
         enable_sleep_mode: bool = False,
         enable_prefix_caching: bool = False,
         enforce_eager: bool = False,