Update services/vision/service.py

services/vision/service.py

@@ -159,9 +159,11 @@ async def _florence_caption(image: Image.Image) -> str:
 
     try:
         device = get_device()
+        dtype = get_torch_dtype()
         _florence_model.to(device)
         prompt = "<MORE_DETAILED_CAPTION>"
         inputs = _florence_processor(text=prompt, images=image, return_tensors="pt").to(device)
+        inputs = {k: v.to(dtype) if v.is_floating_point() else v for k, v in inputs.items()}
 
         with torch.no_grad():
             generated_ids = _florence_model.generate(
@@ -188,9 +190,11 @@ async def _florence_detect(image: Image.Image) -> list[dict]:
 
     try:
         device = get_device()
+        dtype = get_torch_dtype()
         _florence_model.to(device)
         prompt = "<OD>"
         inputs = _florence_processor(text=prompt, images=image, return_tensors="pt").to(device)
+        inputs = {k: v.to(dtype) if v.is_floating_point() else v for k, v in inputs.items()}
 
         with torch.no_grad():
             generated_ids = _florence_model.generate(
@@ -249,7 +253,7 @@ async def _grounding_dino_detect(image: Image.Image) -> list[dict]:
         results = _dino_processor.post_process_grounded_object_detection(
             outputs,
             inputs.input_ids,
-            threshold=VISION_CONFIG["grounding_dino_box_threshold"],
+            box_threshold=VISION_CONFIG["grounding_dino_box_threshold"],
             text_threshold=VISION_CONFIG["grounding_dino_text_threshold"],
             target_sizes=[image.size[::-1]],
         )