[Frontend] [Doc] Exclude log deltas feature (#30322)

Signed-off-by: Catacomba <kevinsuc16@gmail.com>
Signed-off-by: Kevin Šuc <kevinsuc16@gmail.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>

vllm/entrypoints/openai/api_server.py

@@ -1091,6 +1091,7 @@ async def init_app_state(
             enable_prompt_tokens_details=args.enable_prompt_tokens_details,
             enable_force_include_usage=args.enable_force_include_usage,
             enable_log_outputs=args.enable_log_outputs,
+            exclude_log_deltas=args.exclude_log_deltas,
             log_error_stack=args.log_error_stack,
         )
         if "generate" in supported_tasks

vllm/entrypoints/openai/cli_args.py

@@ -187,6 +187,9 @@ class FrontendArgs:
     enable_log_outputs: bool = False
     """If True, log model outputs (generations).
     Requires --enable-log-requests."""
+    exclude_log_deltas: bool = False
+    """If True, model outputs will be logged once streaming is complete. Deltas
+    will not be logged. Requires --enable-log-outputs."""
     h11_max_incomplete_event_size: int = H11_MAX_INCOMPLETE_EVENT_SIZE_DEFAULT
     """Maximum size (bytes) of an incomplete HTTP event (header or body) for
     h11 parser. Helps mitigate header abuse. Default: 4194304 (4 MB)."""
@@ -305,6 +308,8 @@ def validate_parsed_serve_args(args: argparse.Namespace):
     # Enable auto tool needs a tool call parser to be valid
     if args.enable_auto_tool_choice and not args.tool_call_parser:
         raise TypeError("Error: --enable-auto-tool-choice requires --tool-call-parser")
+    if args.exclude_log_deltas and not args.enable_log_outputs:
+        raise TypeError("Error: --exclude-log-deltas requires --enable-log-outputs")
     if args.enable_log_outputs and not args.enable_log_requests:
         raise TypeError("Error: --enable-log-outputs requires --enable-log-requests")
 

vllm/entrypoints/openai/serving_chat.py

@@ -101,6 +101,7 @@ class OpenAIServingChat(OpenAIServing):
         enable_prompt_tokens_details: bool = False,
         enable_force_include_usage: bool = False,
         enable_log_outputs: bool = False,
+        exclude_log_deltas: bool = False,
         log_error_stack: bool = False,
         default_chat_template_kwargs: dict[str, Any] | None = None,
     ) -> None:
@@ -118,6 +119,7 @@ class OpenAIServingChat(OpenAIServing):
         self.trust_request_chat_template = trust_request_chat_template
         self.default_chat_template_kwargs = default_chat_template_kwargs or {}
         self.enable_log_outputs = enable_log_outputs
+        self.exclude_log_deltas = exclude_log_deltas
 
         # set up logits processors
         self.logits_processors = self.model_config.logits_processors
@@ -1135,7 +1137,7 @@ class OpenAIServingChat(OpenAIServing):
                                 if tc.function and tc.function.arguments
                             )
 
-                        if delta_content:
+                        if delta_content and not self.exclude_log_deltas:
                             self.request_logger.log_outputs(
                                 request_id=request_id,
                                 outputs=delta_content,