Update app.py

app.py

@@ -1,64 +1,80 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+import time
+import subprocess
+import os
+from llama_cpp import Llama
+from huggingface_hub import snapshot_download
 
-"""
-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
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+# 下载并转换模型
+def setup_model(model_id):
+    local_dir = model_id.split('/')[-1]
+    if not os.path.exists(local_dir):
+        snapshot_download(repo_id=model_id, local_dir=local_dir)
+    
+    # 转换为 GGUF 格式
+    gguf_path = f"{local_dir}.gguf"
+    if not os.path.exists(gguf_path):
+        subprocess.run(f'python llama.cpp/convert_hf_to_gguf.py ./{local_dir} --outfile {gguf_path}', shell=True, check=True)
+    
+    # 量化模型
+    quantized_path = f"{local_dir}-Q2_K.gguf"
+    if not os.path.exists(quantized_path):
+        subprocess.run(f'./llama.cpp/build/bin/llama-quantize ./{gguf_path} {quantized_path} Q2_K', shell=True, check=True)
+    
+    return quantized_path
 
+# 设定模型路径
+MODEL_ID = "ibm-granite/granite-3.1-2b-instruct"
+MODEL_PATH = setup_model(MODEL_ID)
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+# 加载 Llama 模型
+llm = Llama(
+    model_path=MODEL_PATH,
+    verbose=False,
+    n_threads=4,  # 调整线程数
+    n_ctx=32768   # 上下文窗口大小
+)
 
+def chat_with_model(message, history, system_prompt, temperature, max_tokens, top_k, top_p):
+    """调用 Llama 模型生成回复"""
+    start_time = time.time()
+    
+    messages = [{"role": "system", "content": system_prompt}]
+    for user_msg, assistant_msg in history:
+        messages.append({"role": "user", "content": user_msg})
+        messages.append({"role": "assistant", "content": assistant_msg})
     messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
+    
+    stream = llm.create_chat_completion(
+        messages=messages,
         stream=True,
         temperature=temperature,
+        top_k=top_k,
         top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-
-        response += token
-        yield response
+        max_tokens=max_tokens,
+        stop=["<|im_end|>"]
+    )
+    
+    response = ""
+    for chunk in stream:
+        if "choices" in chunk and chunk["choices"]:
+            text = chunk["choices"][0].get("delta", {}).get("content", "")
+            response += text
+            yield response  # 流式返回文本
 
+    print(f"生成耗时: {time.time() - start_time:.2f} 秒")
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
+# 启动 Gradio ChatInterface
+gr.ChatInterface(
+    fn=chat_with_model,
+    title="Llama GGUF Chatbot",
+    description="使用 Llama GGUF 量化模型进行推理",
+    additional_inputs_accordion=gr.Accordion(label="⚙️ 参数设置", open=False),
     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)",
-        ),
+        gr.Textbox("You are a helpful assistant.", label="System Prompt"),
+        gr.Slider(0, 1, 0.6, label="Temperature"),
+        gr.Slider(100, 4096, 1000, label="Max Tokens"),
+        gr.Slider(1, 100, 40, label="Top K"),
+        gr.Slider(0, 1, 0.85, label="Top P"),
     ],
-)
-
-
-if __name__ == "__main__":
-    demo.launch()
+).queue().launch()