+ app.py

app.py

@@ -0,0 +1,94 @@
+import json
+import os
+import requests
+import gradio as gr
+
+# Environment variables for backend URL and model name
+BACKEND_URL = os.getenv('BACKEND_URL','')
+MODEL_NAME = os.getenv('MODEL_NAME')
+API_KEY = os.getenv('API_KEY')
+
+# Custom headers for the API request
+HEADERS = {
+    'orionstar-api-key': API_KEY,
+    'Content-Type': 'application/json'
+}
+
+def clear_session():
+    """Clears the chat session."""
+    return '', None
+
+def chat_stream_generator(url, payload):
+    """Generator function to stream chat responses from the backend."""
+    answer = ''
+    with requests.post(url, json=payload, headers=HEADERS, stream=True) as response:
+        if response.encoding is None:
+            response.encoding = 'utf-8'
+        for line in response.iter_lines(decode_unicode=True):
+            if line:
+                line = line.replace('data: ', '')
+                if line != '[DONE]':
+                    data = json.loads(line)
+                    if 'choices' in data and data['choices']:
+                        choice = data['choices'][0]
+                        if 'delta' in choice and choice['delta'].get('content'):
+                            answer += choice['delta']['content']
+                            yield answer
+
+def generate_chat(input_text: str, history=None):
+    """Generates chat responses and updates the chat history."""
+    if input_text is None:
+        input_text = ''
+    if history is None:
+        history = []
+    history = history[-5:]  # Keep the last 5 messages in history
+    url = BACKEND_URL
+    payload = {
+        "model": MODEL_NAME,
+        "stream": True,
+        "messages": [
+            {"role": "system", "content": "你是一个AI助手，请用专业的知识解答我的问题"},
+            {"role": "user", "content": input_text}
+        ]
+    }
+
+    gen = chat_stream_generator(url, payload)
+    for response in gen:
+        history.append((input_text, response))
+        yield None, history
+        history.pop()
+    history.append((input_text, response))
+    return None, gen
+
+# Gradio interface
+block = gr.Blocks()
+
+with block as demo:
+    gr.Markdown("<center><h1>猎户星空-Yi-34B-Chat对话模型演示</h1></center>")
+    gr.Markdown("""
+    * Yi系列模型是由零一万物团队开源的大模型，在多个权威的中文、英文及通用领域 benchmark 上取得不错的效果。
+    * [猎户星空](https://www.orionstar.com/)推出的Orionstar-Yi-34B-Chat更进一步挖掘了Yi-34B的潜力。通过对大量高质量微调语料库的深度训练，我们致力于将其打造成为ChatGPT领域中的杰出开源替代品。
+    * Orionstar-Yi-34B-Chat，在C-Eval、MMLU、CMMLU等主流测评集上表现出色，显著领先于国内其他开源对话模型（截止2023年11月）。与其他开源模型的详细对比请查看[这里](https://github.com/OrionStarAI/OrionStar-Yi-34B-Chat)。
+    * 请在[Gihub](https://github.com/OrionStarAI/OrionStar-Yi-34B-Chat)点击Star支持我们。
+    * 加入[官方微信交流群](https://cdn.jsdelivr.net/gh/OrionStarAI/OrionStar-Yi-34B-Chat@main/pics/wechat_group.jpg)""")
+
+    chatbot = gr.Chatbot(label='OrionStar-Yi-34B-Chat', elem_classes="control-height")
+    message = gr.Textbox(label='Input')
+
+    with gr.Row():
+        submit = gr.Button("🚀 Submit")
+        clear_history = gr.Button("🧹 Clear History")
+
+    submit.click(
+        fn=generate_chat,
+        inputs=[message, chatbot],
+        outputs=[message, chatbot]
+    )
+    clear_history.click(
+        fn=clear_session,
+        inputs=[],
+        outputs=[message, chatbot],
+        queue=False
+    )
+
+demo.queue(api_open=False).launch(server_name='0.0.0.0', height=800, share=False,show_api=False)