[DLLM] feat: Add threshold based parallel decoding support (#14412)

Co-authored-by: Jinwei Yao <jinweiy@illinois.edu>
Co-authored-by: 赵晨阳 <zhaochen20@outlook.com>

python/sglang/srt/server_args.py

@@ -461,7 +461,7 @@ class ServerArgs:
 
     # Diffusion LLM
     dllm_algorithm: Optional[str] = None
-    dllm_block_size: Optional[int] = None
+    dllm_algorithm_config: Optional[str] = None
 
     # Double Sparsity
     enable_double_sparsity: bool = False
@@ -3352,13 +3352,13 @@ class ServerArgs:
             "--dllm-algorithm",
             type=str,
             default=ServerArgs.dllm_algorithm,
-            help="The diffusion LLM algorithm.",
+            help="The diffusion LLM algorithm, such as LowConfidence.",
         )
         parser.add_argument(
-            "--dllm-block-size",
-            type=int,
-            default=ServerArgs.dllm_block_size,
-            help="The number of tokens processed in each iteration of the block diffusion LLM.",
+            "--dllm-algorithm-config",
+            type=str,
+            default=ServerArgs.dllm_algorithm_config,
+            help="The diffusion LLM algorithm configurations. Must be a YAML file.",
         )
 
         # Double Sparsity