[None][fix] Fix request_id for best_of/n case (#8368)

Signed-off-by: Olya Kozlova <[email protected]>

Author: evezhier

examples/llm-api/quickstart_advanced.py

@@ -303,7 +303,7 @@ def setup_llm(args, **kwargs):
             args.n = args.max_beam_width
         assert best_of <= args.max_beam_width, f"beam width: {best_of}, should be less or equal to max_beam_width: {args.max_beam_width}"
 
-    assert best_of >= args.n, f"In sampling mode best_of value: {best_of} should be less or equal to n: {args.n}"
+    assert best_of >= args.n, f"In sampling mode best_of value: {best_of} should be greater than or equal to n: {args.n}"
 
     sampling_params = SamplingParams(
         max_tokens=args.max_tokens,

tensorrt_llm/_torch/pyexecutor/llm_request.py

@@ -651,7 +651,7 @@ def create_response(self,
 
         return LlmResponse(
             request_id=self.py_request_id
-            if self.is_child else self.parent_request_id,
+            if not self.is_child else self.parent_request_id,
             result=LlmResult(result, py_result, is_final),
             client_id=self.py_client_id) if len(result) > 0 else None