[None][fix] api stability bug in status label (NVIDIA#7861)

Signed-off-by: Yan Chunwei <[email protected]>
Signed-off-by: Wangshanshan <[email protected]>

Author: Superjomn

tensorrt_llm/llmapi/llm_args.py

@@ -1296,8 +1296,8 @@ class BaseLlmArgs(StrictBaseModel):
     fail_fast_on_attention_window_too_large: bool = Field(
         default=False,
         description=
-        "Fail fast when attention window is too large to fit even a single sequence in the KV cache."
-    )
+        "Fail fast when attention window is too large to fit even a single sequence in the KV cache.",
+        status="prototype")
 
     # LoRA arguments
     enable_lora: bool = Field(default=False, description="Enable LoRA.")

tests/unittest/api_stability/api_stability_core.py

@@ -510,7 +510,9 @@ def check_status(field_name, reference_status, context=""):
             # step 1: check the method status
             method = getattr(self.TEST_CLASS, method_name)
             if method_name in committed_data.get('methods', {}):
-                continue
+                if method_name != "__init__":
+                    continue
+                # Both committed and non-committed methods have __init__ with different parameters
             if method_name != "__init__":
                 method_status = get_api_status(method)
                 if method_status is None: