Update app.py

app.py

@@ -1,49 +1,69 @@
+import spaces  # ⚠️ DEVE SER O PRIMEIRO - Importante para ZeroGPU!
+
 import gradio as gr
 import torch
 from transformers import AutoConfig, AutoModelForCausalLM
 from janus.models import MultiModalityCausalLM, VLChatProcessor
 from janus.utils.io import load_pil_images
 from PIL import Image
-
 import numpy as np
 import os
 import time
 from Upsample import RealESRGAN
-import spaces  # Import spaces for ZeroGPU compatibility
-
 
-# Load model and processor
-model_path = "deepseek-ai/Janus-Pro-7B"
-config = AutoConfig.from_pretrained(model_path)
-language_config = config.language_config
-language_config._attn_implementation = 'eager'
-vl_gpt = AutoModelForCausalLM.from_pretrained(model_path,
-                                             language_config=language_config,
-                                             trust_remote_code=True)
-if torch.cuda.is_available():
-    vl_gpt = vl_gpt.to(torch.bfloat16).cuda()
-else:
-    vl_gpt = vl_gpt.to(torch.float16)
+# Variáveis globais (serão inicializadas dentro das funções)
+vl_gpt = None
+vl_chat_processor = None
+tokenizer = None
+sr_model = None
+model_loaded = False
 
-vl_chat_processor = VLChatProcessor.from_pretrained(model_path)
-tokenizer = vl_chat_processor.tokenizer
-cuda_device = 'cuda' if torch.cuda.is_available() else 'cpu'
-
-# SR model
-sr_model = RealESRGAN(torch.device('cuda' if torch.cuda.is_available() else 'cpu'), scale=2)
-sr_model.load_weights(f'weights/RealESRGAN_x2.pth', download=False)
+def load_models():
+    """Carrega os modelos apenas uma vez, quando necessário"""
+    global vl_gpt, vl_chat_processor, tokenizer, sr_model, model_loaded
+    
+    if model_loaded:
+        return
+    
+    # Carregar modelo principal
+    model_path = "deepseek-ai/Janus-Pro-7B"
+    config = AutoConfig.from_pretrained(model_path)
+    language_config = config.language_config
+    language_config._attn_implementation = 'eager'
+    vl_gpt = AutoModelForCausalLM.from_pretrained(model_path,
+                                                 language_config=language_config,
+                                                 trust_remote_code=True)
+    
+    # Mover para GPU se disponível (sem variável global de device)
+    if torch.cuda.is_available():
+        vl_gpt = vl_gpt.to(torch.bfloat16).cuda()
+    else:
+        vl_gpt = vl_gpt.to(torch.float16)
+    
+    vl_chat_processor = VLChatProcessor.from_pretrained(model_path)
+    tokenizer = vl_chat_processor.tokenizer
+    
+    # Carregar modelo de upscaling
+    device = 'cuda' if torch.cuda.is_available() else 'cpu'
+    sr_model = RealESRGAN(torch.device(device), scale=2)
+    sr_model.load_weights(f'weights/RealESRGAN_x2.pth', download=False)
+    
+    model_loaded = True
 
 @torch.inference_mode()
-@spaces.GPU(duration=120) 
-# Multimodal Understanding function
+@spaces.GPU(duration=120)
 def multimodal_understanding(image, question, seed, top_p, temperature, progress=gr.Progress(track_tqdm=True)):
+    # Garantir que os modelos estão carregados
+    load_models()
+    
     # Clear CUDA cache before generating
     torch.cuda.empty_cache()
     
     # set seed
     torch.manual_seed(seed)
     np.random.seed(seed)
-    torch.cuda.manual_seed(seed)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed(seed)
     
     conversation = [
         {
@@ -55,10 +75,13 @@ def multimodal_understanding(image, question, seed, top_p, temperature, progress
     ]
     
     pil_images = [Image.fromarray(image)]
+    
+    device = 'cuda' if torch.cuda.is_available() else 'cpu'
+    dtype = torch.bfloat16 if torch.cuda.is_available() else torch.float16
+    
     prepare_inputs = vl_chat_processor(
         conversations=conversation, images=pil_images, force_batchify=True
-    ).to(cuda_device, dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float16)
-    
+    ).to(device, dtype=dtype)
     
     inputs_embeds = vl_gpt.prepare_inputs_embeds(**prepare_inputs)
     
@@ -78,7 +101,6 @@ def multimodal_understanding(image, question, seed, top_p, temperature, progress
     answer = tokenizer.decode(outputs[0].cpu().tolist(), skip_special_tokens=True)
     return answer
 
-
 def generate(input_ids,
              width,
              height,
@@ -88,16 +110,19 @@ def generate(input_ids,
              image_token_num_per_image: int = 576,
              patch_size: int = 16,
              progress=gr.Progress(track_tqdm=True)):
+    
+    device = 'cuda' if torch.cuda.is_available() else 'cpu'
+    
     # Clear CUDA cache before generating
     torch.cuda.empty_cache()
     
-    tokens = torch.zeros((parallel_size * 2, len(input_ids)), dtype=torch.int).to(cuda_device)
+    tokens = torch.zeros((parallel_size * 2, len(input_ids)), dtype=torch.int).to(device)
     for i in range(parallel_size * 2):
         tokens[i, :] = input_ids
         if i % 2 != 0:
             tokens[i, 1:-1] = vl_chat_processor.pad_id
     inputs_embeds = vl_gpt.language_model.get_input_embeddings()(tokens)
-    generated_tokens = torch.zeros((parallel_size, image_token_num_per_image), dtype=torch.int).to(cuda_device)
+    generated_tokens = torch.zeros((parallel_size, image_token_num_per_image), dtype=torch.int).to(device)
 
     pkv = None
     for i in range(image_token_num_per_image):
@@ -119,8 +144,6 @@ def generate(input_ids,
             img_embeds = vl_gpt.prepare_gen_img_embeds(next_token)
             inputs_embeds = img_embeds.unsqueeze(dim=1)
 
-    
-
     patches = vl_gpt.gen_vision_model.decode_code(generated_tokens.to(dtype=torch.int),
                                                  shape=[parallel_size, 8, width // patch_size, height // patch_size])
 
@@ -135,22 +158,27 @@ def unpack(dec, width, height, parallel_size=5):
 
     return visual_img
 
-
-
 @torch.inference_mode()
-@spaces.GPU(duration=120)  # Specify a duration to avoid timeout
+@spaces.GPU(duration=120)
 def generate_image(prompt,
                    seed=None,
                    guidance=5,
                    t2i_temperature=1.0,
                    progress=gr.Progress(track_tqdm=True)):
+    
+    # Garantir que os modelos estão carregados
+    load_models()
+    
     # Clear CUDA cache and avoid tracking gradients
     torch.cuda.empty_cache()
+    
     # Set the seed for reproducible results
     if seed is not None:
         torch.manual_seed(seed)
-        torch.cuda.manual_seed(seed)
+        if torch.cuda.is_available():
+            torch.cuda.manual_seed(seed)
         np.random.seed(seed)
+    
     width = 384
     height = 384
     parallel_size = 4
@@ -175,18 +203,17 @@ def generate_image(prompt,
                         height // 16 * 16,
                         parallel_size=parallel_size)
 
-        # return [Image.fromarray(images[i]).resize((768, 768), Image.LANCZOS) for i in range(parallel_size)]
-        stime = time.time()
         ret_images = [image_upsample(Image.fromarray(images[i])) for i in range(parallel_size)]
-        print(f'upsample time: {time.time() - stime}')
         return ret_images
 
-
 @spaces.GPU(duration=60)
 def image_upsample(img: Image.Image) -> Image.Image:
     if img is None:
         raise Exception("Image not uploaded")
     
+    # Garantir que o modelo de upscaling está carregado
+    load_models()
+    
     width, height = img.size
     
     if width >= 5000 or height >= 5000:
@@ -195,14 +222,15 @@ def image_upsample(img: Image.Image) -> Image.Image:
     global sr_model
     result = sr_model.predict(img.convert('RGB'))
     return result
-        
 
 # Gradio interface
 css = '''
 .gradio-container {max-width: 960px !important}
 '''
+
 with gr.Blocks(css=css) as demo:
     gr.Markdown("# Janus Pro 7B")
+    
     with gr.Tab("Multimodal Understanding"):
         gr.Markdown(value="## Multimodal Understanding")
         image_input = gr.Image()
@@ -213,29 +241,14 @@ with gr.Blocks(css=css) as demo:
         understanding_output = gr.Textbox(label="Response")
         
         with gr.Accordion("Advanced options", open=False):
-                und_seed_input = gr.Number(label="Seed", precision=0, value=42)
-                top_p = gr.Slider(minimum=0, maximum=1, value=0.95, step=0.05, label="top_p")
-                temperature = gr.Slider(minimum=0, maximum=1, value=0.1, step=0.05, label="temperature")
-            
-        examples_inpainting = gr.Examples(
-            label="Multimodal Understanding examples",
-            examples=[
-                [
-                    "explain this meme",
-                    "doge.png",
-                ],
-                [
-                    "Convert the formula into latex code.",
-                    "equation.png",
-                ],
-            ],
-            inputs=[question_input, image_input],
-        )
+            und_seed_input = gr.Number(label="Seed", precision=0, value=42)
+            top_p = gr.Slider(minimum=0, maximum=1, value=0.95, step=0.05, label="top_p")
+            temperature = gr.Slider(minimum=0, maximum=1, value=0.1, step=0.05, label="temperature")
     
     with gr.Tab("Text-to-Image Generation"):
         gr.Markdown(value="## Text-to-Image Generation")
 
-        prompt_input = gr.Textbox(label="Prompt. (Prompt in more detail can help produce better images!")
+        prompt_input = gr.Textbox(label="Prompt. (Prompt in more detail can help produce better images!)")
     
         generation_button = gr.Button("Generate Images")
     
@@ -246,7 +259,7 @@ with gr.Blocks(css=css) as demo:
                 cfg_weight_input = gr.Slider(minimum=1, maximum=10, value=5, step=0.5, label="CFG Weight")
                 t2i_temperature = gr.Slider(minimum=0, maximum=1, value=1.0, step=0.05, label="temperature")
             seed_input = gr.Number(label="Seed (Optional)", precision=0, value=1234)
-    
+        
         examples_t2i = gr.Examples(
             label="Text to image generation examples.",
             examples=[
@@ -254,7 +267,6 @@ with gr.Blocks(css=css) as demo:
                 "The face of a beautiful girl",
                 "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
                 "A cute and adorable baby fox with big brown eyes, autumn leaves in the background enchanting,immortal,fluffy, shiny mane,Petals,fairyism,unreal engine 5 and Octane Render,highly detailed, photorealistic, cinematic, natural colors.",
-                "The image features an intricately designed eye set against a circular backdrop adorned with ornate swirl patterns that evoke both realism and surrealism. At the center of attention is a strikingly vivid blue iris surrounded by delicate veins radiating outward from the pupil to create depth and intensity. The eyelashes are long and dark, casting subtle shadows on the skin around them which appears smooth yet slightly textured as if aged or weathered over time.\n\nAbove the eye, there's a stone-like structure resembling part of classical architecture, adding layers of mystery and timeless elegance to the composition. This architectural element contrasts sharply but harmoniously with the organic curves surrounding it. Below the eye lies another decorative motif reminiscent of baroque artistry, further enhancing the overall sense of eternity encapsulated within each meticulously crafted detail. \n\nOverall, the atmosphere exudes a mysterious aura intertwined seamlessly with elements suggesting timelessness, achieved through the juxtaposition of realistic textures and surreal artistic flourishes. Each component\u2014from the intricate designs framing the eye to the ancient-looking stone piece above\u2014contributes uniquely towards creating a visually captivating tableau imbued with enigmatic allure.",
             ],
             inputs=prompt_input,
         )