Update app.py

app.py

@@ -168,13 +168,13 @@ pipe = load_and_prepare_model()
 #checkpoint = "microsoft/Phi-3.5-mini-instruct"
 checkpoint = "ford442/Phi-3.5-mini-instruct-bf16"
 #captioner = pipeline(model="ydshieh/vit-gpt2-coco-en",device='cuda:0', task="image-to-text")
-captioner_2 = pipeline(model="Salesforce/blip-image-captioning-base",device='cuda', task="image-to-text")
-#captioner_3 = pipeline(model="ford442/blip-image-to-text-large-bf16",device='cuda', task="image-to-text")
-model5 = Blip2ForConditionalGeneration.from_pretrained("ford442/blip2-image-to-text-bf16").to('cuda')
+#captioner_2 = pipeline(model="Salesforce/blip-image-captioning-base",device='cuda', task="image-to-text")
+captioner_2 = pipeline(model="ford442/blip-image-to-text-large-bf16",device='cuda', task="image-to-text")
+model5 = Blip2ForConditionalGeneration.from_pretrained("ford442/blip2-image-to-text-bf16").to('cuda', torch.bfloat16)
 processor5 = Blip2Processor.from_pretrained("ford442/blip2-image-to-text-bf16", device_map='cuda')
-#txt_tokenizer = AutoTokenizer.from_pretrained(checkpoint, device_map='cuda', add_prefix_space=False)
-#txt_tokenizer.tokenizer_legacy=False
-#model = Phi3ForCausalLM.from_pretrained(checkpoint).to('cuda:0')
+txt_tokenizer = AutoTokenizer.from_pretrained(checkpoint, device_map='cuda', add_prefix_space=False)
+txt_tokenizer.tokenizer_legacy=False
+model = Phi3ForCausalLM.from_pretrained(checkpoint).to('cuda:0',torch.bfloat16)
 #model = AutoModelForCausalLM.from_pretrained(checkpoint, device_map='cuda') #.to('cuda')
 
 ip_model = IPAdapterXL(pipe, local_folder, ip_ckpt, device)
@@ -435,8 +435,8 @@ def generate_30(
         gc.collect()
         torch.cuda.empty_cache()
         torch.cuda.reset_peak_memory_stats()
-        #expanded = expand_prompt(captions)
-        new_prompt = captions #prompt+' '+expanded
+        expanded = expand_prompt(captions)
+        new_prompt = expanded
         print("-- ------------ --")
         print("-- FINAL PROMPT --")
         print(new_prompt)
@@ -588,8 +588,8 @@ def generate_60(
         gc.collect()
         torch.cuda.empty_cache()
         torch.cuda.reset_peak_memory_stats()
-        #expanded = expand_prompt(captions)
-        new_prompt = captions #prompt+' '+expanded
+        expanded = expand_prompt(captions)
+        new_prompt = expanded
         print("-- ------------ --")
         print("-- FINAL PROMPT --")
         print(new_prompt)
@@ -741,8 +741,8 @@ def generate_90(
         gc.collect()
         torch.cuda.empty_cache()
         torch.cuda.reset_peak_memory_stats()
-        #expanded = expand_prompt(captions)
-        new_prompt = captions #prompt+' '+expanded
+        expanded = expand_prompt(captions)
+        new_prompt = expanded
         print("-- ------------ --")
         print("-- FINAL PROMPT --")
         print(new_prompt)