Update app.py

app.py

@@ -1,64 +1,138 @@
+
+import spaces  # 必须在最顶部导入
 import gradio as gr
-from huggingface_hub import InferenceClient
+import os
+
+# 获取 Hugging Face 访问令牌
+hf_token = os.getenv("HF_API_TOKEN")
+
+# 定义基础模型名称
+base_model_name = "larry1129/WooWoof_AI_Vision_merged_16bit_3b"
+
+# 定义 adapter 模型名称
+# adapter_model_name = "larry1129/WooWoof_AI"
 
+# 定义全局变量用于缓存模型和分词器
+model = None
+tokenizer = None
+
+# 定义提示生成函数
+def generate_prompt(instruction, input_text=""):
+    if input_text:
+        prompt = f"""### Instruction:
+{instruction}
+### Input:
+{input_text}
+### Response:
 """
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
+    else:
+        prompt = f"""### Instruction:
+{instruction}
+### Response:
 """
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+    return prompt
 
+# 定义生成响应的函数，并使用 @spaces.GPU 装饰
+@spaces.GPU(duration=40)  # 建议将 duration 增加到 120
+def generate_response(instruction, input_text):
+    global model, tokenizer
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+    if model is None:
+        print("开始加载模型...")
+        # 检查 bitsandbytes 是否已安装
+        import importlib.util
+        if importlib.util.find_spec("bitsandbytes") is None:
+            import subprocess
+            subprocess.call(["pip", "install", "--upgrade", "bitsandbytes"])
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+        try:
+            # 在函数内部导入需要 GPU 的库
+            import torch
+            from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
 
-    messages.append({"role": "user", "content": message})
+            from peft import PeftModel
 
-    response = ""
+            # 创建量化配置
+            bnb_config = BitsAndBytesConfig(
+                load_in_4bit=True,
+                bnb_4bit_use_double_quant=True,
+                bnb_4bit_quant_type="nf4",
+                bnb_4bit_compute_dtype=torch.float16
+            )
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+            # 加载分词器
+            tokenizer = AutoTokenizer.from_pretrained(base_model_name, use_auth_token=hf_token)
+            print("分词器加载成功。")
 
-        response += token
-        yield response
+            # 加载基础模型
+            base_model = AutoModelForCausalLM.from_pretrained(
+                base_model_name,
+                quantization_config=bnb_config,
+                device_map="auto",
+                use_auth_token=hf_token,
+                trust_remote_code=True
+            )
+            print("基础模型加载成功。")
 
+            # 加载适配器模型
+            model = PeftModel.from_pretrained(
+                base_model,
+                 # adapter_model_name,
+                torch_dtype=torch.float16,
+                use_auth_token=hf_token
+            )
+            print("适配器模型加载成功。")
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
+            # 设置 pad_token
+            tokenizer.pad_token = tokenizer.eos_token
+            model.config.pad_token_id = tokenizer.pad_token_id
+
+            # 切换到评估模式
+            model.eval()
+            print("模型已切换到评估模式。")
+        except Exception as e:
+            print("加载模型时出错：", e)
+            raise e
+    else:
+        # 在函数内部导入需要的库
+        import torch
+
+    # 检查 model 和 tokenizer 是否已正确加载
+    if model is None or tokenizer is None:
+        print("模型或分词器未正确加载。")
+        raise ValueError("模型或分词器未正确加载。")
+
+    # 生成提示
+    prompt = generate_prompt(instruction, input_text)
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+
+    with torch.no_grad():
+        outputs = model.generate(
+            input_ids=inputs["input_ids"],
+            attention_mask=inputs.get("attention_mask"),
+            max_new_tokens=128,
+            temperature=0.7,
+            top_p=0.95,
+            do_sample=True,
+        )
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    response = response.split("### Response:")[-1].strip()
+    return response
+
+# 创建 Gradio 接口
+iface = gr.Interface(
+    fn=generate_response,
+    inputs=[
+        gr.Textbox(lines=2, placeholder="Instruction", label="Instruction"),
+        gr.Textbox(lines=2, placeholder="Input", label="Input (Option)")
     ],
+    outputs="text",
+    title="WooWoof AI",
+    description="Based on LLAMA 3.1 for pet related",
+    allow_flagging="never"
 )
 
+# 启动 Gradio 接口
+iface.launch(share=True)
+
 
-if __name__ == "__main__":
-    demo.launch()