API: Fix error reporting

Make a disconnect on load error consistently. It should be safer to
warn the user to run unload (or re-run load) if a model does not
load correctly.

Also don't log the traceback for request errors that don't have one.

Signed-off-by: kingbri <[email protected]>

common/utils.py

@@ -27,15 +27,15 @@ class TabbyRequestError(BaseModel):
     error: TabbyRequestErrorMessage
 
 
-def get_generator_error(message: str):
+def get_generator_error(message: str, exc_info: bool = True):
     """Get a generator error."""
 
     generator_error = handle_request_error(message)
 
     return get_sse_packet(generator_error.model_dump_json())
 
 
-def handle_request_error(message: str):
+def handle_request_error(message: str, exc_info: bool = True):
     """Log a request error to the console."""
 
     error_message = TabbyRequestErrorMessage(
@@ -45,7 +45,7 @@ def handle_request_error(message: str):
     request_error = TabbyRequestError(error=error_message)
 
     # Log the error and provided message to the console
-    if error_message.trace:
+    if error_message.trace and exc_info:
         logger.error(error_message.trace)
 
     logger.error(f"Sent to request: {message}")

main.py

@@ -93,11 +93,14 @@
 
 
 async def _check_model_container():
+    """Checks if a model isn't loading or loaded."""
+
     if MODEL_CONTAINER is None or not (
         MODEL_CONTAINER.model_is_loading or MODEL_CONTAINER.model_loaded
     ):
         error_message = handle_request_error(
-            "No models are currently loaded."
+            "No models are currently loaded.",
+            exc_info=False,
         ).error.message
 
         raise HTTPException(400, error_message)
@@ -221,6 +224,7 @@ async def generator():
 
         # Unload the existing model
         if MODEL_CONTAINER and MODEL_CONTAINER.model:
+            logger.info("Unloading existing model.")
             await unload_model()
 
         MODEL_CONTAINER = ExllamaV2Container(model_path.resolve(), False, **load_data)
@@ -231,7 +235,10 @@ async def generator():
         try:
             for module, modules in load_status:
                 if await request.is_disconnected():
-                    await unload_model()
+                    logger.error(
+                        "Model load cancelled by user. "
+                        "Please make sure to run unload to free up resources."
+                    )
                     break
 
                 if module == 0: