feat(model-runtime): thread request_metadata through to model runtimes

src/graphon/model_runtime/callbacks/base_callback.py

@@ -37,7 +37,7 @@ def on_before_invoke(
         stop: Sequence[str] | None = None,
         stream: bool = True,
         user: str | None = None,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
     ) -> None:
         """Before invoke callback
 
@@ -50,7 +50,7 @@ def on_before_invoke(
         :param stop: stop words
         :param stream: is stream response
         :param user: optional end-user identifier for the invocation
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         """
         raise NotImplementedError
 
@@ -67,7 +67,7 @@ def on_new_chunk(
         stop: Sequence[str] | None = None,
         stream: bool = True,
         user: str | None = None,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
     ) -> None:
         """On new chunk callback
 
@@ -81,7 +81,7 @@ def on_new_chunk(
         :param stop: stop words
         :param stream: is stream response
         :param user: optional end-user identifier for the invocation
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         """
         raise NotImplementedError
 
@@ -98,7 +98,7 @@ def on_after_invoke(
         stop: Sequence[str] | None = None,
         stream: bool = True,
         user: str | None = None,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
     ) -> None:
         """After invoke callback
 
@@ -112,7 +112,7 @@ def on_after_invoke(
         :param stop: stop words
         :param stream: is stream response
         :param user: optional end-user identifier for the invocation
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         """
         raise NotImplementedError
 
@@ -129,7 +129,7 @@ def on_invoke_error(
         stop: Sequence[str] | None = None,
         stream: bool = True,
         user: str | None = None,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
     ) -> None:
         """Invoke error callback
 
@@ -143,7 +143,7 @@ def on_invoke_error(
         :param stop: stop words
         :param stream: is stream response
         :param user: optional end-user identifier for the invocation
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         """
         raise NotImplementedError
 

src/graphon/model_runtime/callbacks/logging_callback.py

@@ -28,7 +28,7 @@ def on_before_invoke(
         stop: Sequence[str] | None = None,
         stream: bool = True,
         user: str | None = None,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
     ) -> None:
         """Before invoke callback
 
@@ -41,7 +41,7 @@ def on_before_invoke(
         :param stop: stop words
         :param stream: is stream response
         :param user: optional end-user identifier for the invocation
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         """
         self.print_text("\n[on_llm_before_invoke]\n", color="blue")
         self.print_text(f"Model: {model}\n", color="blue")
@@ -61,9 +61,9 @@ def on_before_invoke(
         if user:
             self.print_text(f"User: {user}\n", color="blue")
 
-        if invocation_context:
+        if request_metadata:
             self.print_text(
-                f"Invocation context: {dict(invocation_context)}\n",
+                f"Request metadata: {dict(request_metadata)}\n",
                 color="blue",
             )
 
@@ -91,7 +91,7 @@ def on_new_chunk(
         stop: Sequence[str] | None = None,
         stream: bool = True,
         user: str | None = None,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
     ) -> None:
         """On new chunk callback
 
@@ -104,9 +104,9 @@ def on_new_chunk(
         :param tools: tools for tool calling
         :param stop: stop words
         :param stream: is stream response
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         """
-        _ = user, invocation_context
+        _ = user, request_metadata
         if isinstance(chunk.delta.message.content, str):
             sys.stdout.write(chunk.delta.message.content)
             sys.stdout.flush()
@@ -124,7 +124,7 @@ def on_after_invoke(
         stop: Sequence[str] | None = None,
         stream: bool = True,
         user: str | None = None,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
     ) -> None:
         """After invoke callback
 
@@ -137,9 +137,9 @@ def on_after_invoke(
         :param tools: tools for tool calling
         :param stop: stop words
         :param stream: is stream response
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         """
-        _ = user, invocation_context
+        _ = user, request_metadata
         self.print_text("\n[on_llm_after_invoke]\n", color="yellow")
         self.print_text(f"Content: {result.message.content}\n", color="yellow")
 
@@ -173,7 +173,7 @@ def on_invoke_error(
         stop: Sequence[str] | None = None,
         stream: bool = True,
         user: str | None = None,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
     ) -> None:
         """Invoke error callback
 
@@ -186,8 +186,8 @@ def on_invoke_error(
         :param tools: tools for tool calling
         :param stop: stop words
         :param stream: is stream response
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         """
-        _ = user, invocation_context
+        _ = user, request_metadata
         self.print_text("\n[on_llm_invoke_error]\n", color="red")
         logger.error("LLM invoke failed: %s", ex, exc_info=ex)

src/graphon/model_runtime/model_providers/base/large_language_model.py

@@ -284,6 +284,7 @@ def _invoke_llm_via_runtime(
     tools: list[PromptMessageTool] | None,
     stop: Sequence[str] | None,
     stream: bool,
+    request_metadata: Mapping[str, object] | None,
 ) -> LLMResult | Generator[LLMResultChunk, None, None]:
     return llm_model.model_runtime.invoke_llm(
         provider=provider,
@@ -294,6 +295,7 @@ def _invoke_llm_via_runtime(
         tools=tools,
         stop=stop,
         stream=stream,
+        request_metadata=request_metadata,
     )
 
 
@@ -312,6 +314,8 @@ def invoke(
         stop: list[str] | None = None,
         stream: bool = True,
         callbacks: list[Callback] | None = None,
+        *,
+        request_metadata: Mapping[str, object] | None = None,
     ) -> LLMResult | Generator[LLMResultChunk, None, None]:
         """Invoke the large language model and optionally stream result chunks."""
         # validate and filter model parameters
@@ -334,6 +338,7 @@ def invoke(
             tools=tools,
             stop=stop,
             stream=stream,
+            request_metadata=request_metadata,
             callbacks=callbacks,
         )
 
@@ -350,6 +355,7 @@ def invoke(
                 tools=tools,
                 stop=stop,
                 stream=stream,
+                request_metadata=request_metadata,
             )
 
             if not stream:
@@ -368,6 +374,7 @@ def invoke(
                 tools=tools,
                 stop=stop,
                 stream=stream,
+                request_metadata=request_metadata,
                 callbacks=callbacks,
             )
 
@@ -383,6 +390,7 @@ def invoke(
                 tools=tools,
                 stop=stop,
                 stream=stream,
+                request_metadata=request_metadata,
                 callbacks=callbacks,
             )
         if isinstance(result, LLMResult):
@@ -395,6 +403,7 @@ def invoke(
                 tools=tools,
                 stop=stop,
                 stream=stream,
+                request_metadata=request_metadata,
                 callbacks=callbacks,
             )
             # Following https://github.com/langgenius/dify/issues/17799,
@@ -415,7 +424,7 @@ def _invoke_result_generator(
         tools: list[PromptMessageTool] | None = None,
         stop: Sequence[str] | None = None,
         stream: bool = True,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
         callbacks: list[Callback] | None = None,
     ) -> Generator[LLMResultChunk, None, None]:
         """Stream runtime result chunks through callbacks and bookkeeping hooks."""
@@ -433,7 +442,7 @@ def _invoke_result_generator(
                 tools=tools,
                 stop=stop,
                 stream=stream,
-                invocation_context=invocation_context,
+                request_metadata=request_metadata,
                 callbacks=callbacks,
             )
         except Exception as e:
@@ -450,7 +459,7 @@ def _invoke_result_generator(
             tools=tools,
             stop=stop,
             stream=stream,
-            invocation_context=invocation_context,
+            request_metadata=request_metadata,
             callbacks=callbacks,
         )
 
@@ -466,7 +475,7 @@ def _stream_invoke_result_chunks(
         tools: list[PromptMessageTool] | None,
         stop: Sequence[str] | None,
         stream: bool,
-        invocation_context: Mapping[str, object] | None,
+        request_metadata: Mapping[str, object] | None,
         callbacks: list[Callback],
     ) -> Generator[LLMResultChunk, None, None]:
         for chunk in result:
@@ -486,7 +495,7 @@ def _stream_invoke_result_chunks(
                 tools=tools,
                 stop=stop,
                 stream=stream,
-                invocation_context=invocation_context,
+                request_metadata=request_metadata,
                 callbacks=callbacks,
             )
             accumulator.consume(chunk)
@@ -558,7 +567,7 @@ def _trigger_before_invoke_callbacks(
         tools: list[PromptMessageTool] | None = None,
         stop: Sequence[str] | None = None,
         stream: bool = True,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
         callbacks: list[Callback] | None = None,
     ) -> None:
         """Trigger before invoke callbacks
@@ -570,7 +579,7 @@ def _trigger_before_invoke_callbacks(
         :param tools: tools for tool calling
         :param stop: stop words
         :param stream: is stream response
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         :param callbacks: callbacks
         """
         _run_callbacks(
@@ -585,7 +594,7 @@ def _trigger_before_invoke_callbacks(
                 tools=tools,
                 stop=stop,
                 stream=stream,
-                invocation_context=invocation_context,
+                request_metadata=request_metadata,
             ),
         )
 
@@ -599,7 +608,7 @@ def _trigger_new_chunk_callbacks(
         tools: list[PromptMessageTool] | None = None,
         stop: Sequence[str] | None = None,
         stream: bool = True,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
         callbacks: list[Callback] | None = None,
     ) -> None:
         """Trigger new chunk callbacks
@@ -612,7 +621,7 @@ def _trigger_new_chunk_callbacks(
         :param tools: tools for tool calling
         :param stop: stop words
         :param stream: is stream response
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         """
         _run_callbacks(
             callbacks,
@@ -627,7 +636,7 @@ def _trigger_new_chunk_callbacks(
                 tools=tools,
                 stop=stop,
                 stream=stream,
-                invocation_context=invocation_context,
+                request_metadata=request_metadata,
             ),
         )
 
@@ -641,7 +650,7 @@ def _trigger_after_invoke_callbacks(
         tools: list[PromptMessageTool] | None = None,
         stop: Sequence[str] | None = None,
         stream: bool = True,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
         callbacks: list[Callback] | None = None,
     ) -> None:
         """Trigger after invoke callbacks
@@ -654,7 +663,7 @@ def _trigger_after_invoke_callbacks(
         :param tools: tools for tool calling
         :param stop: stop words
         :param stream: is stream response
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         :param callbacks: callbacks
         """
         _run_callbacks(
@@ -670,7 +679,7 @@ def _trigger_after_invoke_callbacks(
                 tools=tools,
                 stop=stop,
                 stream=stream,
-                invocation_context=invocation_context,
+                request_metadata=request_metadata,
             ),
         )
 
@@ -684,7 +693,7 @@ def _trigger_invoke_error_callbacks(
         tools: list[PromptMessageTool] | None = None,
         stop: Sequence[str] | None = None,
         stream: bool = True,
-        invocation_context: Mapping[str, object] | None = None,
+        request_metadata: Mapping[str, object] | None = None,
         callbacks: list[Callback] | None = None,
     ) -> None:
         """Trigger invoke error callbacks
@@ -697,7 +706,7 @@ def _trigger_invoke_error_callbacks(
         :param tools: tools for tool calling
         :param stop: stop words
         :param stream: is stream response
-        :param invocation_context: opaque request metadata for the current invocation
+        :param request_metadata: opaque metadata for the current request
         :param callbacks: callbacks
         """
         _run_callbacks(
@@ -713,6 +722,6 @@ def _trigger_invoke_error_callbacks(
                 tools=tools,
                 stop=stop,
                 stream=stream,
-                invocation_context=invocation_context,
+                request_metadata=request_metadata,
             ),
         )