Update app.py

app.py

@@ -243,6 +243,7 @@ def uploadNote(prompt,num_inference_steps,guidance_scale,timestamp):
         f.write(f"Model UNET: ford442/RealVisXL_V5.0_BF16 \n")
     upload_to_ftp(filename) 
 
+@torch.no_grad()
 def captioning(img):
     prompts_array = [
        # "Adjectives describing this scene are:",
@@ -264,7 +265,7 @@ def captioning(img):
         **inputsa,
         do_sample=False,
         num_beams=5,
-        max_length=96,
+        max_length=512,
         #min_length=1,
         top_p=0.9,
         repetition_penalty=1.5,
@@ -278,17 +279,19 @@ def captioning(img):
     # Loop through prompts array:
     for prompt in prompts_array:
         inputs = processor5(images=img, text=prompt, return_tensors="pt").to('cuda')
+        
         generated_ids = model5.generate(
             **inputs,
             do_sample=False,
             num_beams=5,
-            max_length=128,
-            min_length=42,
+            max_length=256,
+            #min_length=42,
             top_p=0.9,
             repetition_penalty=1.5,
             length_penalty=1.0,
             temperature=1,
         ) 
+        
         generated_text = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
         response_text = generated_text.replace(prompt, "").strip() #Or could try .split(prompt, 1)[-1].strip()
         output_prompt.append(response_text)
@@ -296,7 +299,7 @@ def captioning(img):
         
             # Continue conversation:
     inputf = processor5(images=img, text=generated_text + 'So therefore', return_tensors="pt").to('cuda')
-    generated_ids = model5.generate(**inputf, max_length=192)
+    generated_ids = model5.generate(**inputf, max_length=768)
     generated_text = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
     response_text = generated_text.replace(generated_text, "").strip()
     print(response_text)
@@ -434,7 +437,7 @@ def generate_30(
         timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
         filename= f'rv_IPb_{timestamp}.png'
         print("-- using image file --")
-        captions =caption.append(prompt)
+        captions = caption.append(flatten_and_stringify(prompt))
         captions = flatten_and_stringify(captions)
         captions = " ".join(captions)
         print(captions)