Update app.py

app.py

@@ -25,7 +25,7 @@ blip_processor_large = AutoProcessor.from_pretrained("Salesforce/blip-image-capt
 blip_model_large = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")
 
 blip2_processor = AutoProcessor.from_pretrained("Salesforce/blip2-opt-2.7b")
-blip2_model = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-2.7b")
+blip2_model = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-2.7b", torch_dtype=torch.float16)
 
 # vitgpt_processor = AutoImageProcessor.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
 # vitgpt_model = VisionEncoderDecoderModel.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
@@ -47,8 +47,11 @@ blip_model_large.to(device)
 coca_model.to(device)
 blip2_model.to(device)
 
-def generate_caption(processor, model, image, tokenizer=None):
+def generate_caption(processor, model, image, tokenizer=None, use_float_16=False):
     inputs = processor(images=image, return_tensors="pt").to(device)
+
+    if use_float_16:
+        inputs = inputs.to(torch.float16)
     
     generated_ids = model.generate(pixel_values=inputs.pixel_values, max_length=50)
 
@@ -82,7 +85,7 @@ def generate_captions(image):
 
     caption_coca = generate_caption_coca(coca_model, coca_transform, image)
 
-    caption_blip2 = generate_caption(blip2_processor, blip2_model, image)
+    caption_blip2 = generate_caption(blip2_processor, blip2_model, image, use_float_16=True)
 
     return caption_git_large_coco, caption_git_large_textcaps, caption_blip_large, caption_coca, caption_blip2