app.py

app.py

@@ -1,13 +1,17 @@
 from transformers import AutoProcessor, AutoModelForVision2Seq
 from PIL import Image
-import torch
 import gradio as gr
+import torch
 
 MODEL_ID = "HuggingFaceM4/idefics2-8b"
 
-# Load model and processor
+# 强制使用 CPU 模式
+device = "cpu"
+
+# 加载模型与处理器（关闭 float16 避免 CPU 报错）
 processor = AutoProcessor.from_pretrained(MODEL_ID)
-model = AutoModelForVision2Seq.from_pretrained(MODEL_ID, torch_dtype=torch.float16, device_map="auto")
+model = AutoModelForVision2Seq.from_pretrained(MODEL_ID, torch_dtype=torch.float32, device_map=None)
+model.to(device)
 
 def analyze_images(base_img, target_img, user_prompt):
     if base_img is None or target_img is None:
@@ -16,9 +20,8 @@ def analyze_images(base_img, target_img, user_prompt):
     images = [base_img, target_img]
     prompt = f"Ignore the first image (base image). Analyze the second image: {user_prompt}"
     
-    inputs = processor(images=images, text=prompt, return_tensors="pt").to(model.device)
+    inputs = processor(images=images, text=prompt, return_tensors="pt").to(device)
     output = model.generate(**inputs, max_new_tokens=200)
-    
     result = processor.decode(output[0], skip_special_tokens=True)
     return result
 
@@ -30,8 +33,8 @@ demo = gr.Interface(
         gr.Textbox(label="Prompt", placeholder="Describe what to analyze...")
     ],
     outputs=gr.Textbox(label="Model Output"),
-    title="Image Comparison with IDEFICS2-8B",
-    description="Upload two images. The model will ignore the base image and analyze the target image according to your prompt."
+    title="Image Comparison (IDEFICS2-8B, CPU Mode)",
+    description="Upload two images. The model will ignore the base image and analyze the second according to your prompt."
 )
 
 if __name__ == "__main__":