Add vllm id to the response

src/guidellm/backends/response_handlers.py

@@ -116,6 +116,7 @@ def compile_non_streaming(
             request_args=str(
                 request.arguments.model_dump() if request.arguments else None
             ),
+            response_id=response.get("id"), # use vLLM ID if available
             text=choices[0].get("text", "") if choices else "",
             input_metrics=input_metrics,
             output_metrics=output_metrics,

src/guidellm/schemas/response.py

@@ -41,6 +41,10 @@ class GenerationResponse(StandardBaseModel):
     request_id: str = Field(
         description="Unique identifier matching the original GenerationRequest."
     )
+    response_id: str | None = Field(
+        default=None,
+        description="Unique identifier matching the original vLLM Response ID."
+    )
     request_args: str | None = Field(
         description="Arguments passed to the backend for request processing."
     )
@@ -108,6 +112,7 @@ def compile_stats(
 
         return GenerativeRequestStats(
             request_id=self.request_id,
+            response_id=self.response_id,
             request_type=request.request_type,
             request_args=str(
                 request.arguments.model_dump() if request.arguments else {}

src/guidellm/schemas/stats.py

@@ -45,6 +45,9 @@ class GenerativeRequestStats(StandardBaseDict):
     request_type: GenerativeRequestType | str = Field(
         description="Type of generative request: text or chat completion"
     )
+    response_id: str | None = Field(
+        default=None, description="vLLM-assigned unique request ID"
+    )
     request_args: str | None = Field(
         default=None, description="Arguments passed to the backend for this request"
     )