Update app.py

app.py

@@ -1,58 +1,66 @@
 import gradio as gr
 import numpy as np
 import random
-import spaces
 import torch
-from diffusers import  DiffusionPipeline, FlowMatchEulerDiscreteScheduler, AutoencoderTiny, AutoencoderKL
-from transformers import CLIPTextModel, CLIPTokenizer,T5EncoderModel, T5TokenizerFast
-from live_preview_helpers import calculate_shift, retrieve_timesteps, flux_pipe_call_that_returns_an_iterable_of_images
+from diffusers import DiffusionPipeline, AutoencoderTiny, AutoencoderKL
+from transformers import CLIPTextModel, CLIPTokenizer
+from live_preview_helpers import flux_pipe_call_that_returns_an_iterable_of_images
 
-dtype = torch.bfloat16
+# Definindo variáveis e carregando modelos
+dtype = torch.float16  # Usando float16 para melhorar a performance
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
 taef1 = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype).to(device)
 good_vae = AutoencoderKL.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="vae", torch_dtype=dtype).to(device)
 pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=dtype, vae=taef1).to(device)
+pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(pipe)
+
 torch.cuda.empty_cache()
 
 MAX_SEED = np.iinfo(np.int32).max
 MAX_IMAGE_SIZE = 2048
 
-pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(pipe)
-
+# Função de inferência otimizada
+@torch.inference_mode()  # Desabilitando cálculo de gradientes para acelerar a inferência
 @spaces.GPU(duration=75)
 def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=3.5, num_inference_steps=28, progress=gr.Progress(track_tqdm=True)):
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)
-    generator = torch.Generator().manual_seed(seed)
     
-    for img in pipe.flux_pipe_call_that_returns_an_iterable_of_images(
-            prompt=prompt,
-            guidance_scale=guidance_scale,
-            num_inference_steps=num_inference_steps,
-            width=width,
-            height=height,
-            generator=generator,
-            output_type="pil",
-            good_vae=good_vae,
-        ):
+    generator = torch.Generator(device).manual_seed(seed)
+    
+    # Usando autograd em precisão reduzida (float16) para acelerar a inferência
+    with torch.autocast("cuda", dtype=torch.float16):
+        for img in pipe.flux_pipe_call_that_returns_an_iterable_of_images(
+                prompt=prompt,
+                guidance_scale=guidance_scale,
+                num_inference_steps=num_inference_steps,
+                width=width,
+                height=height,
+                generator=generator,
+                output_type="pil",
+                good_vae=good_vae,
+            ):
             yield img, seed
     
+    torch.cuda.empty_cache()  # Limpar a memória após a inferência para liberar recursos
+
+# Exemplos
 examples = [
     "a tiny astronaut hatching from an egg on the moon",
     "a cat holding a sign that says hello world",
     "an anime illustration of a wiener schnitzel",
 ]
 
-css="""
+css = """
 #col-container {
     margin: 0 auto;
     max-width: 520px;
 }
 """
 
+# Interface Gradio
 with gr.Blocks(css=css) as demo:
-    
     with gr.Column(elem_id="col-container"):
         gr.Markdown(f"""# FLUX.1 [dev]
 12B param rectified flow transformer guidance-distilled from [FLUX.1 [pro]](https://blackforestlabs.ai/)  
@@ -60,7 +68,6 @@ with gr.Blocks(css=css) as demo:
         """)
         
         with gr.Row():
-            
             prompt = gr.Text(
                 label="Prompt",
                 show_label=False,
@@ -68,13 +75,11 @@ with gr.Blocks(css=css) as demo:
                 placeholder="Enter your prompt",
                 container=False,
             )
-            
             run_button = gr.Button("Run", scale=0)
         
         result = gr.Image(label="Result", show_label=False)
         
         with gr.Accordion("Advanced Settings", open=False):
-            
             seed = gr.Slider(
                 label="Seed",
                 minimum=0,
@@ -86,7 +91,6 @@ with gr.Blocks(css=css) as demo:
             randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
             
             with gr.Row():
-                
                 width = gr.Slider(
                     label="Width",
                     minimum=256,
@@ -104,7 +108,6 @@ with gr.Blocks(css=css) as demo:
                 )
             
             with gr.Row():
-
                 guidance_scale = gr.Slider(
                     label="Guidance Scale",
                     minimum=1,
@@ -122,18 +125,18 @@ with gr.Blocks(css=css) as demo:
                 )
         
         gr.Examples(
-            examples = examples,
-            fn = infer,
-            inputs = [prompt],
-            outputs = [result, seed],
+            examples=examples,
+            fn=infer,
+            inputs=[prompt],
+            outputs=[result, seed],
             cache_examples="lazy"
         )
 
     gr.on(
         triggers=[run_button.click, prompt.submit],
-        fn = infer,
-        inputs = [prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
-        outputs = [result, seed]
+        fn=infer,
+        inputs=[prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
+        outputs=[result, seed]
     )
 
-demo.launch()
+demo.launch()