Update app.py

app.py

@@ -246,14 +246,14 @@ def captioning(img):
     output_prompt=[]
     # Initial caption generation without a prompt:
     inputsa = processor5(images=img, return_tensors="pt").to('cuda')
-    generated_ids = model5.generate(**inputsa, min_length=32, max_length=64)
+    generated_ids = model5.generate(**inputsa, min_length=42, max_length=64)
     generated_text = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
     output_prompt.append(generated_text)
     print(generated_text)  
     # 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) # Adjust max_length if needed
+        generated_ids = model5.generate(**inputs, min_length=16, max_length=64) # Adjust max_length if needed
         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)
@@ -291,7 +291,7 @@ def expand_prompt(prompt):
         outputs = model.generate(
             input_ids=input_ids,
             attention_mask=attention_mask,
-            max_new_tokens=1024,
+            max_new_tokens=512,
             temperature=0.2,
             top_p=0.9,
             do_sample=True,
@@ -304,7 +304,7 @@ def expand_prompt(prompt):
         outputs_2 = model.generate(
             input_ids=input_ids_2,
             attention_mask=attention_mask_2,
-            max_new_tokens=1024,
+            max_new_tokens=512,
             temperature=0.2,
             top_p=0.9,
             do_sample=True,