Update app.py

app.py

@@ -15,13 +15,9 @@ from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
 import gradio as gr
 from accelerate import Accelerator
 
-
 # Instantiate the Accelerator
 accelerator = Accelerator()
 
-# Use the accelerator to prepare your models, optimizers, and data loaders
-model, optimizer, dataloader = accelerator.prepare(model, optimizer, dataloader)
-
 dtype = torch.bfloat16
 
 # Set environment variables for local path
@@ -62,10 +58,10 @@ output_dir = 'generated_images'
 os.makedirs(output_dir, exist_ok=True)
 
 # Function to generate a detailed visual description prompt
-def generate_description_prompt(subject, user_prompt):
-    prompt = f"write concise vivid visual description enclosed in brackets like [ <description> ] less than 100 words of {user_prompt} different from {subject}. "
+def generate_description_prompt(subject, user_prompt, text_generator):
+    prompt = f"write concise vivid visual description enclosed in brackets like [ <description> ] less than 50 words of {user_prompt} different from {subject}. "
     try:
-        generated_text = text_generator(prompt, max_length=230, num_return_sequences=1, truncation=True)[0]['generated_text']
+        generated_text = text_generator(prompt, max_length=160, num_return_sequences=1, truncation=True)[0]['generated_text']
         generated_description = re.sub(rf'{re.escape(prompt)}\s*', '', generated_text).strip()  # Remove the prompt from the generated text
         return generated_description if generated_description else None
     except Exception as e:
@@ -115,7 +111,7 @@ def generate_and_store_descriptions(user_prompt, batch_size=100, max_iterations=
             break
 
         subject = random.choice(available_subjects)
-        generated_description = generate_description_prompt(subject, user_prompt)
+        generated_description = generate_description_prompt(subject, user_prompt, text_generator)
         
         if generated_description:
             # Remove any offending symbols