use smaller model to improve CUDA usage

app.py

@@ -166,4 +166,4 @@ with gr.Blocks() as demo:
         ]
     )
 
-demo.launch(debug=False, show_error=True, max_threads=1)
+demo.launch(debug=False, show_error=True)

utils/florence.py

@@ -7,7 +7,7 @@ from PIL import Image
 from transformers import AutoModelForCausalLM, AutoProcessor
 from transformers.dynamic_module_utils import get_imports
 
-FLORENCE_CHECKPOINT = "microsoft/Florence-2-large"
+FLORENCE_CHECKPOINT = "microsoft/Florence-2-base"
 FLORENCE_OBJECT_DETECTION_TASK = '<OD>'
 FLORENCE_DETAILED_CAPTION_TASK = '<MORE_DETAILED_CAPTION>'
 FLORENCE_CAPTION_TO_PHRASE_GROUNDING_TASK = '<CAPTION_TO_PHRASE_GROUNDING>'

utils/sam.py

@@ -7,8 +7,8 @@ from PIL import Image
 from sam2.build_sam import build_sam2
 from sam2.sam2_image_predictor import SAM2ImagePredictor
 
-SAM_CHECKPOINT = "checkpoints/sam2_hiera_large.pt"
-SAM_CONFIG = "sam2_hiera_l.yaml"
+SAM_CHECKPOINT = "checkpoints/sam2_hiera_small.pt"
+SAM_CONFIG = "sam2_hiera_s.yaml"
 
 
 def load_sam_model(