Update app.py

app.py

@@ -1,130 +1,93 @@
-import gradio as gr
-from gradio_client import Client
-from PIL import Image
 import os
-import time
-import traceback
-import asyncio
-from huggingface_hub import HfApi
-
-# Your Hugging Face API key (ensure this is set in your environment or replace directly)
-api_key = os.getenv('MY_API_KEY')
-api = HfApi(token=api_key)
-
-# List of repos (private spaces)
-repos = [
-    "hsuwill000/LCM-absolutereality-openvino-8bit_00",    
-    "hsuwill000/LCM-absolutereality-openvino-8bit_01",
-    "hsuwill000/LCM-absolutereality-openvino-8bit_02",
-    "hsuwill000/LCM-absolutereality-openvino-8bit_03",
-    "hsuwill000/LCM-absolutereality-openvino-8bit_04",
-    "hsuwill000/LCM-absolutereality-openvino-8bit_05",    
-    "hsuwill000/LCM-absolutereality-openvino-8bit_06",
-    "hsuwill000/LCM-absolutereality-openvino-8bit_07",
-    "hsuwill000/LCM-absolutereality-openvino-8bit_08",
-    "hsuwill000/LCM-absolutereality-openvino-8bit_09",
-    "hsuwill000/LCM-absolutereality-openvino-8bit_10",
-]
-
-class CustomClient(Client):
-    def __init__(self, *args, timeout=30, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.timeout = timeout
-
-    def _request(self, method, url, **kwargs):
-        kwargs['timeout'] = self.timeout
-        return super()._request(method, url, **kwargs)
+import gradio as gr
+from transformers import AutoModelForCausalLM
+from optimum.intel.openvino import OVStableDiffusionPipeline
+import torch
 
-# Counter for image filenames to avoid overwriting
-count = 0
+# 定義模型 ID 與存儲路徑
+model_id = "Kouki79/Realistic_Vision6_LCM"
+export_path = "exported_model_openvino_int8"
 
-async def infer_single_gradio(client, prompt):
-    global count
-    # Prepare the inputs for the prediction
-    inputs = {
-        "prompt": prompt,
-        #"num_inference_steps": 10  # Number of inference steps for the model
-    }
+# 設定圖片大小
+HIGH = 1024
+WIDTH = 512
 
-    try:
-        # Send the request to the model and receive the image
-        result = await asyncio.to_thread(client.predict, inputs, api_name="/infer")
-        
-        # Open the resulting image
-        image = Image.open(result)
-        
-        # Create a unique filename to save the image
-        filename = f"img_{count:08d}.jpg"
-        while os.path.exists(filename):
-            count += 1
-            filename = f"img_{count:08d}.jpg"
-        
-        # Save the image locally
-        image.save(filename)
-        print(f"Saved image as {filename}")
-        
-        # Return the image to be displayed in Gradio
-        return image
+print("🔍 檢查 OpenVINO 模型是否已存在...")
+if not os.path.exists(export_path) or not os.listdir(export_path):
+    print("⚠️ 尚未轉換 OpenVINO 8-bit 模型，開始轉換...")
     
-    except Exception as e:
-        # Handle any errors that occur
-        print(f"An exception occurred: {str(e)}")
-        print("Stack trace:")
-        traceback.print_exc()  # Print stack trace for debugging
-        return None  # Return nothing if an error occurs
+    # 轉換 Hugging Face 模型為 OpenVINO 8-bit
+    model = OVStableDiffusionPipeline.from_pretrained(
+        model_id,
+        export=True,  # 自動轉換為 OpenVINO
+        device="CPU",
+        precision="int8",  # 啟用 8-bit 量化
+    )
 
-async def infer_gradio(prompt: str, output_images: gr.Gallery):
-    #can't work Clear previous images from the gallery
-    #output_images.update([])
+    # 儲存轉換後的 OpenVINO 8-bit 模型
+    model.save_pretrained(export_path)
+    print(f"✅ 轉換完成！OpenVINO 8-bit 模型已儲存至 '{export_path}'")
+else:
+    print(f"✅ 發現已轉換的 OpenVINO 8-bit 模型：'{export_path}'")
 
-    # Record the start time
-    start_time = time.time()
+# 載入 OpenVINO 8-bit 模型
+print("🔄 載入 OpenVINO 8-bit 模型...")
+pipe = OVStableDiffusionPipeline.from_pretrained(
+    export_path,
+    compile=True,  # 編譯模型以提高效能
+    device="CPU",
+    safety_checker=None,
+    torch_dtype=torch.uint8
+)
+print("✅ OpenVINO 模型載入完成！")
 
-    # Create a list of tasks (one for each repo)
-    tasks = []
-    for repo in repos:
-        runtime_info = api.get_space_runtime(repo)
-        if runtime_info.stage == 'APP_STARTING':
-            continue
-        if runtime_info.stage in ['SLEEPING', 'PAUSED']:
-            print(f"{repo} is now SLEEPING or PAUSED.")
-            api.restart_space(repo_id=repo)
-            continue   
-        # Create a CustomClient instance for each repo
-        client = CustomClient(repo, hf_token=api_key, timeout=300)
-        task = infer_single_gradio(client, prompt)
-        tasks.append(task)
+# 設定推論函數
+def infer(prompt):
+    print(f"🖼️ 生成圖片: {prompt}")
+    image = pipe(
+        prompt=f",hyper-realistic 2K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic,",
+        negative_prompt="EasyNegative, cartoonish, low resolution, blurry, simplistic, abstract, deformed, ugly,",
+        width=WIDTH,
+        height=HIGH,
+        guidance_scale=1.0,
+        num_inference_steps=6,
+        num_images_per_prompt=1,
+    ).images[0]
     
-    # Run all tasks concurrently (i.e., generate images from all repos)
-    results = await asyncio.gather(*tasks)
+    return image
 
-    # Calculate the time taken for image generation
-    end_time = time.time()
-    time_taken = end_time - start_time
-    print(f"Time taken to generate the image(s): {time_taken:.2f} seconds")
+# Gradio UI 設定
+css = """
+#col-container {
+    margin: 0 auto;
+    max-width: 520px;
+}
+"""
 
-    # Return the results (images)
-    return results  # Return all the images as a list
-
-# Define Gradio Interface
-with gr.Blocks() as demo:
-    with gr.Row():  # Use a Row to place the prompt input and the button side by side
-        prompt_input = gr.Textbox(
-            label="Enter Your Prompt", 
-            show_label="False",
-            placeholder="Type your prompt for image generation here",
-            lines=1,  # Set the input to be only one line tall
-            interactive=True  # Allow user to interact with the textbox
-        )
+with gr.Blocks(css=css) as demo:
+    with gr.Column(elem_id="col-container"):
+        gr.Markdown(f"""
+        # {model_id.split('/')[1]} {WIDTH}x{HIGH}
+        Running on OpenVINO (8-bit).
+        """)
         
-        # Change the button text to "RUN:" and align it with the prompt input
-        run_button = gr.Button("RUN")
-    
-    # Output image display area (will show multiple images)
-    output_images = gr.Gallery(label="Generated Images", elem_id="gallery", show_label=False)
-    
-    # Connecting the button click to the image generation function
-    run_button.click(infer_gradio, inputs=[prompt_input], outputs=output_images)
+        with gr.Row():
+            prompt = gr.Textbox(
+                label="Prompt",
+                show_label=False,
+                max_lines=1,
+                placeholder="Enter your prompt",
+                container=False,
+            )         
+            run_button = gr.Button("Generate", scale=0)
+        
+        result = gr.Image(label="Result", show_label=False)
+
+    run_button.click(
+        fn=infer,
+        inputs=[prompt],
+        outputs=[result]
+    )
 
-# Launch Gradio app
-demo.launch()
+print("🚀 啟動 Gradio Web UI...")
+demo.queue().launch()