Update app.py

app.py

@@ -1,49 +1,42 @@
 import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
+from transformers import AutoTokenizer, AutoModelForVision2Seq
 import gradio as gr
+from PIL import Image
 
-model_name = "deepseek-ai/deepseek-vl-1.3b-chat"
+# 模型名称
+model_id = "deepseek-ai/deepseek-vl-1.3b-chat"
 
-# 加载 tokenizer 和模型
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
-    device_map="auto" if torch.cuda.is_available() else None
-)
-model.generation_config = GenerationConfig.from_pretrained(model_name)
-model.generation_config.pad_token_id = model.generation_config.eos_token_id
+# 加载 tokenizer 和 model
+tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
+model = AutoModelForVision2Seq.from_pretrained(model_id, torch_dtype=torch.float16, trust_remote_code=True).to("cuda")
+model.eval()
+
+# 图文聊天函数
+def chat_with_image(image: Image.Image, user_input: str):
+    # 构造 prompt
+    messages = [
+        {"role": "user", "content": [
+            {"type": "image", "image": image},
+            {"type": "text", "text": user_input}
+        ]}
+    ]
 
-# 推理函数
-def chat(prompt, max_new_tokens=256, temperature=0.7, top_p=0.9):
-    messages = [{"role": "user", "content": prompt}]
-    input_tensor = tokenizer.apply_chat_template(
-        messages, add_generation_prompt=True, return_tensors="pt"
-    ).to(model.device)
+    # 使用 generate_response 方法（根据 DeepSeek 的源码）
     with torch.no_grad():
-        outputs = model.generate(
-            input_tensor,
-            max_new_tokens=max_new_tokens,
-            temperature=temperature,
-            top_p=top_p,
-            do_sample=True,
-            eos_token_id=tokenizer.eos_token_id
-        )
-    result = tokenizer.decode(outputs[0][input_tensor.shape[1]:], skip_special_tokens=True)
-    return result
+        output = model.chat(tokenizer, messages=messages, image=image)
+
+    return output
 
-# Gradio 界面
+# Gradio 接口
 iface = gr.Interface(
-    fn=chat,
+    fn=chat_with_image,
     inputs=[
-        gr.Textbox(lines=4, label="请输入您的问题"),
-        gr.Slider(32, 1024, step=16, value=256, label="最大生成长度"),
-        gr.Slider(0.1, 1.5, step=0.1, value=0.7, label="Temperature"),
-        gr.Slider(0.1, 1.0, step=0.05, value=0.9, label="Top-p")
+        gr.Image(type="pil", label="上传图片"),
+        gr.Textbox(label="请输入你的问题")
     ],
-    outputs=gr.Textbox(label="模型回复"),
-    title="DeepSeekChat 演示",
-    description="基于 Hugging Face Spaces 的部署示例"
+    outputs=gr.Textbox(label="模型回答"),
+    title="DeepSeek-VL-1.3B Chat Demo",
+    description="上传图片并输入问题，体验多模态聊天模型。"
 )
 
 if __name__ == "__main__":