Fixup GenerativeRequestSession

Author: sjmonson

src/guidellm/request/session.py

@@ -44,17 +44,18 @@ def __len__(self) -> int:
 
     def get_next_request(self) -> GenerationRequest:
         completed_responses = len(self.responses)
-        base_request = self.prompts[completed_responses].model_copy()
+        base_request = self.prompts[completed_responses].model_copy(deep=True)
         base_request.content = "".join(
             itertools.chain.from_iterable(
-                zip((x.content for x in self.prompts), self.responses)
+                zip((x.content for x in self.prompts), self.responses + [""])
             )
         )
         base_request.stats["prompt_tokens"] = sum(
             x.stats["prompt_tokens"] for x in self.prompts[: completed_responses + 1]
         )
         base_request.constraints["output_tokens"] = sum(
-            x.constraints["output_tokens"] for x in self.prompts[:completed_responses]
+            x.constraints["output_tokens"]
+            for x in self.prompts[: completed_responses + 1]
         )
 
         return base_request