Update app.py

app.py

@@ -1,172 +1,97 @@
 import gradio as gr
 from random import randint
 from all_models import models
-
 from externalmod import gr_Interface_load, randomize_seed
-
 import asyncio
 import os
 from threading import RLock
 
-# Create a lock to ensure thread safety when accessing shared resources
+# Create a lock for thread safety
 lock = RLock()
-# Load Hugging Face token from environment variable, if available
-HF_TOKEN = os.environ.get("HF_TOKEN") if os.environ.get("HF_TOKEN") else None # If private or gated models aren't used, ENV setting is unnecessary.
 
-# Function to load all models specified in the 'models' list
+# Load Hugging Face token from environment variables
+HF_TOKEN = os.environ.get("HF_TOKEN")
+
+# Function to load models
 def load_fn(models):
     global models_load
     models_load = {}
-    
-    # Iterate through all models to load them
     for model in models:
-        if model not in models_load.keys():
+        if model not in models_load:
             try:
-                # Log model loading attempt
-                print(f"Attempting to load model: {model}")
-                # Load model interface using externalmod function
+                print(f"Loading model: {model}")
                 m = gr_Interface_load(f'models/{model}', hf_token=HF_TOKEN)
-                print(f"Successfully loaded model: {model}")
+                print(f"Successfully loaded: {model}")
             except Exception as error:
-                # In case of an error, print it and create a placeholder interface
-                print(f"Error loading model {model}: {error}")
+                print(f"Error loading {model}: {error}")
                 m = gr.Interface(lambda: None, ['text'], ['image'])
-            # Update the models_load dictionary with the loaded model
-            models_load.update({model: m})
+            models_load[model] = m
 
-# Load all models defined in the 'models' list
+# Load models
 print("Loading models...")
 load_fn(models)
 print("Models loaded successfully.")
 
-num_models = 6
-
-# Set the default models to use for inference
-default_models = models[:num_models]
-inference_timeout = 600
-MAX_SEED = 3999999999
-# Generate a starting seed randomly between 1941 and 2024
+num_models = min(3, len(models))  # Reduce to 3 models to prevent GPU overloading
 starting_seed = randint(1941, 2024)
 print(f"Starting seed: {starting_seed}")
 
-# Extend the choices list to ensure it contains 'num_models' elements
+# Extend choices to match num_models
 def extend_choices(choices):
-    print(f"Extending choices: {choices}")
     extended = choices[:num_models] + (num_models - len(choices[:num_models])) * ['NA']
-    print(f"Extended choices: {extended}")
     return extended
 
-# Update the image boxes based on selected models
+# Update image boxes based on selected models
 def update_imgbox(choices):
-    print(f"Updating image boxes with choices: {choices}")
-    choices_plus = extend_choices(choices[:num_models])
-    imgboxes = [gr.Image(None, label=m, visible=(m != 'NA')) for m in choices_plus]
-    print(f"Updated image boxes: {imgboxes}")
-    return imgboxes
+    choices_extended = extend_choices(choices)
+    return [gr.Image(None, label=m, visible=(m != 'NA')) for m in choices_extended]
 
-# Asynchronous function to perform inference on a given model
-async def infer(model_str, prompt, seed=1, timeout=inference_timeout):
-    from pathlib import Path
-    kwargs = {}
-    noise = ""
-    kwargs["seed"] = seed
-    # Create an asynchronous task to run the model inference
-    print(f"Starting inference for model: {model_str} with prompt: '{prompt}' and seed: {seed}")
-    task = asyncio.create_task(asyncio.to_thread(models_load[model_str].fn,
-                               prompt=f'{prompt} {noise}', **kwargs, token=HF_TOKEN))
-    await asyncio.sleep(0)  # Allow other tasks to run
+# Asynchronous inference function
+async def infer(model_str, prompt, seed=1, timeout=600):
+    if model_str == 'NA':
+        return None
     try:
-        # Wait for the task to complete within the specified timeout
+        print(f"Running inference on {model_str} with prompt: '{prompt}'")
+        task = asyncio.to_thread(models_load[model_str].fn, prompt=prompt, seed=seed, token=HF_TOKEN)
         result = await asyncio.wait_for(task, timeout=timeout)
-        print(f"Inference completed for model: {model_str}")
-    except (Exception, asyncio.TimeoutError) as e:
-        # Handle any exceptions or timeout errors
-        print(f"Error during inference for model {model_str}: {e}")
-        if not task.done():
-            task.cancel()
-            print(f"Task cancelled for model: {model_str}")
-        result = None
-    # If the task completed successfully, save the result as an image
-    if task.done() and result is not None:
-        with lock:
-            png_path = "image.png"
-            result.save(png_path)
-            image = str(Path(png_path).resolve())
-            print(f"Result saved as image: {image}")
-        return image
-    print(f"No result for model: {model_str}")
+        if result:
+            with lock:
+                image_path = "image.png"
+                result.save(image_path)
+                return image_path
+    except Exception as e:
+        print(f"Error in inference for {model_str}: {e}")
     return None
 
-# Function to generate an image based on the given model, prompt, and seed
+# Wrapper function for inference
 def gen_fnseed(model_str, prompt, seed=1):
     if model_str == 'NA':
-        print(f"Model is 'NA', skipping generation.")
         return None
-    try:
-        # Create a new event loop to run the asynchronous inference function
-        print(f"Generating image for model: {model_str} with prompt: '{prompt}' and seed: {seed}")
-        loop = asyncio.new_event_loop()
-        result = loop.run_until_complete(infer(model_str, prompt, seed, inference_timeout))
-    except (Exception, asyncio.CancelledError) as e:
-        # Handle any exceptions or cancelled tasks
-        print(f"Error during generation for model {model_str}: {e}")
-        result = None
-    finally:
-        # Close the event loop
-        loop.close()
-        print(f"Event loop closed for model: {model_str}")
-    return result
+    return asyncio.run(infer(model_str, prompt, seed))
 
-# Create the Gradio Blocks interface with a custom theme
+# Create Gradio interface
 print("Creating Gradio interface...")
 with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
-    gr.HTML("<center><h1>Compare-6</h1></center>")
-    with gr.Tab('Compare-6'):
-        # Text input for user prompt
+    gr.HTML("<center><h1>Compare-3</h1></center>")
+    with gr.Tab('Compare-3'):
         txt_input = gr.Textbox(label='Your prompt:', lines=4)
-        # Button to generate images
-        gen_button = gr.Button('Generate up to 6 images in up to 3 minutes total')
-        with gr.Row():
-            # Slider to select a seed for reproducibility
-            seed = gr.Slider(label="Use a seed to replicate the same image later (maximum 3999999999)", minimum=0, maximum=MAX_SEED, step=1, value=starting_seed, scale=3)
-            # Button to randomize the seed
-            seed_rand = gr.Button("Randomize Seed 🎲", size="sm", variant="secondary", scale=1)    
-        # Set up click event to randomize the seed
-        seed_rand.click(randomize_seed, None, [seed], queue=False)
-        print("Seed randomization button set up.")
-        # Button click to start generation
-        gen_button.click(lambda s: gr.update(interactive=True), None)
-        print("Generation button set up.")
-
-        with gr.Row():
-            # Create image output components for each model
-            output = [gr.Image(label=m, min_width=480) for m in default_models]
-            # Create hidden textboxes to store the current models
-            current_models = [gr.Textbox(m, visible=False) for m in default_models]
-            
-            # Set up generation events for each model and output image
-            for m, o in zip(current_models, output):
-                print(f"Setting up generation event for model: {m.value}")
-                gen_event = gr.on(triggers=[gen_button.click, txt_input.submit], fn=gen_fnseed,
-                                  inputs=[m, txt_input, seed], outputs=[o], concurrency_limit=None, queue=False)
-                # The commented stop button could be used to cancel the generation event
-                #stop_button.click(lambda s: gr.update(interactive=False), None, stop_button, cancels=[gen_event])
-        # Accordion to allow model selection
+        gen_button = gr.Button('Generate images')
+        seed = gr.Slider(label="Seed (max 3999999999)", minimum=0, maximum=3999999999, step=1, value=starting_seed)
+        seed_rand = gr.Button("Randomize Seed 🎲")
+        seed_rand.click(randomize_seed, None, [seed])
+        
+        output = [gr.Image(label=m) for m in models[:num_models]]
+        current_models = [gr.Textbox(m, visible=False) for m in models[:num_models]]
+        
+        for m, o in zip(current_models, output):
+            gen_button.click(gen_fnseed, inputs=[m, txt_input, seed], outputs=[o])
+        
         with gr.Accordion('Model selection'):
-            # Checkbox group to select up to 'num_models' different models
-            model_choice = gr.CheckboxGroup(models, label=f'Choose up to {int(num_models)} different models from the {len(models)} available!', value=default_models, interactive=True)
-            # Update image boxes and current models based on model selection
+            model_choice = gr.CheckboxGroup(models, label=f'Choose up to {num_models} models', value=models[:num_models])
             model_choice.change(update_imgbox, model_choice, output)
             model_choice.change(extend_choices, model_choice, current_models)
-            print("Model selection setup complete.")
-        with gr.Row():
-            # Placeholder HTML to add additional UI elements if needed
-            gr.HTML(
-)
 
-# Queue settings for handling multiple concurrent requests
-print("Setting up queue...")
-demo.queue(default_concurrency_limit=200, max_size=200)
+# Reduce concurrency to avoid T4 overload
+demo.queue(default_concurrency_limit=50, max_size=100)
 print("Launching Gradio interface...")
-demo.launch(show_api=False, max_threads=400)
-print("Gradio interface launched successfully.")
+demo.launch(show_api=False, max_threads=50, debug=True)