k

backend/danswer/document_index/vespa/app_config/validation-overrides.xml

@@ -2,4 +2,7 @@
     <allow
         until="DATE_REPLACEMENT"
         comment="We need to be able to create/delete indices for swapping models">schema-removal</allow>
+    <allow
+        until="DATE_REPLACEMENT"
+        comment="We need to be able to update the schema for updates to the Danswer schema">indexing-change</allow>
 </validation-overrides>

backend/model_server/main.py

@@ -54,9 +54,11 @@ def _move_files_recursively(source: Path, dest: Path, overwrite: bool = False) -
 @asynccontextmanager
 async def lifespan(app: FastAPI) -> AsyncGenerator:
     if torch.cuda.is_available():
-        logger.notice("GPU is available")
+        logger.notice("CUDA GPU is available")
+    elif torch.backends.mps.is_available():
+        logger.notice("Mac MPS is available")
     else:
-        logger.notice("GPU is not available")
+        logger.notice("GPU is not available, using CPU")
 
     if TEMP_HF_CACHE_PATH.is_dir():
         logger.notice("Moving contents of temp_huggingface to huggingface cache.")

backend/model_server/management_endpoints.py

@@ -11,6 +11,10 @@ def healthcheck() -> Response:
 
 
 @router.get("/gpu-status")
-def gpu_status() -> dict[str, bool]:
-    has_gpu = torch.cuda.is_available()
-    return {"gpu_available": has_gpu}
+def gpu_status() -> dict[str, bool | str]:
+    if torch.cuda.is_available():
+        return {"gpu_available": True, "type": "cuda"}
+    elif torch.backends.mps.is_available():
+        return {"gpu_available": True, "type": "mps"}
+    else:
+        return {"gpu_available": False, "type": "none"}