SPO / app.py
tang-x's picture
Update app.py
0c752de verified
raw
history blame
14.6 kB
import asyncio
from pathlib import Path
from typing import Dict, List
import streamlit as st
import yaml
from loguru import logger as _logger
import shutil
import uuid
from metagpt.const import METAGPT_ROOT
from metagpt.ext.spo.components.optimizer import PromptOptimizer
from metagpt.ext.spo.utils.llm_client import SPO_LLM, RequestType
def get_user_workspace():
if "user_id" not in st.session_state:
st.session_state.user_id = str(uuid.uuid4())
workspace_dir = Path("workspace") / st.session_state.user_id
workspace_dir.mkdir(parents=True, exist_ok=True)
return workspace_dir
def cleanup_workspace(workspace_dir: Path) -> None:
try:
if workspace_dir.exists():
shutil.rmtree(workspace_dir)
_logger.info(f"Cleaned up workspace directory: {workspace_dir}")
except Exception as e:
_logger.error(f"Error cleaning up workspace: {e}")
def get_template_path(template_name: str, is_new_template: bool = False) -> str:
"""
Get template file path
:param template_name: Name of the template
:param is_new_template: Whether it's a new template created by user
:return: Path object for the template file
"""
if is_new_template:
# Create user-specific subdirectory in settings folder
if "user_id" not in st.session_state:
st.session_state.user_id = str(uuid.uuid4())
user_settings_path = st.session_state.user_id
return f"{user_settings_path}/{template_name}.yaml"
else:
# Use root settings path for existing templates
return f"{template_name}.yaml"
def get_all_templates() -> List[str]:
"""
Get list of all available templates (both default and user-specific)
:return: List of template names
"""
settings_path = Path("metagpt/ext/spo/settings")
# Get default templates
templates = [f.stem for f in settings_path.glob("*.yaml")]
# Get user-specific templates if user_id exists
if "user_id" in st.session_state:
user_path = settings_path / st.session_state.user_id
if user_path.exists():
user_templates = [f"{st.session_state.user_id}/{f.stem}" for f in user_path.glob("*.yaml")]
templates.extend(user_templates)
return sorted(list(set(templates)))
def load_yaml_template(template_path: Path) -> Dict:
if template_path.exists():
with open(template_path, "r", encoding="utf-8") as f:
return yaml.safe_load(f)
return {"prompt": "", "requirements": "", "count": None, "qa": [{"question": "", "answer": ""}]}
def save_yaml_template(template_path: Path, data: Dict, is_new: bool) -> None:
if is_new:
template_format = {
"prompt": str(data.get("prompt", "")),
"requirements": str(data.get("requirements", "")),
"count": data.get("count"),
"qa": [
{"question": str(qa.get("question", "")).strip(), "answer": str(qa.get("answer", "")).strip()}
for qa in data.get("qa", [])
],
}
template_path.parent.mkdir(parents=True, exist_ok=True)
with open(template_path, "w", encoding="utf-8") as f:
yaml.dump(template_format, f, allow_unicode=True, sort_keys=False, default_flow_style=False, indent=2)
else:
pass
def display_optimization_results(result_data):
for result in result_data:
round_num = result["round"]
success = result["succeed"]
prompt = result["prompt"]
with st.expander(f"轮次 {round_num} {':white_check_mark:' if success else ':x:'}"):
st.markdown("**提示词:**")
st.code(prompt, language="text")
st.markdown("<br>", unsafe_allow_html=True)
col1, col2 = st.columns(2)
with col1:
st.markdown(f"**状态:** {'成功 ✅ ' if success else '失败 ❌ '}")
with col2:
st.markdown(f"**令牌数:** {result['tokens']}")
st.markdown("**回答:**")
for idx, answer in enumerate(result["answers"]):
st.markdown(f"**问题 {idx + 1}:**")
st.text(answer["question"])
st.markdown("**答案:**")
st.text(answer["answer"])
st.markdown("---")
# 总结
success_count = sum(1 for r in result_data if r["succeed"])
total_rounds = len(result_data)
st.markdown("### 总结")
col1, col2 = st.columns(2)
with col1:
st.metric("总轮次", total_rounds)
with col2:
st.metric("成功轮次", success_count)
def main():
if "optimization_results" not in st.session_state:
st.session_state.optimization_results = []
workspace_dir = get_user_workspace()
st.markdown(
"""
<div style="background-color: #f0f2f6; padding: 20px; border-radius: 10px; margin-bottom: 25px">
<div style="display: flex; justify-content: space-between; align-items: center; margin-bottom: 10px">
<h1 style="margin: 0;">SPO | 自监督提示词优化 🤖</h1>
</div>
<div style="display: flex; gap: 20px; align-items: center">
<a href="https://arxiv.org/pdf/2502.06855" target="_blank" style="text-decoration: none;">
<img src="https://img.shields.io/badge/论文-PDF-red.svg" alt="论文">
</a>
<a href="https://github.com/geekan/MetaGPT/blob/main/examples/spo/README.md" target="_blank" style="text-decoration: none;">
<img src="https://img.shields.io/badge/GitHub-仓库-blue.svg" alt="GitHub">
</a>
<span style="color: #666;">一个自监督提示词优化框架</span>
</div>
</div>
""",
unsafe_allow_html=True
)
# 侧边栏配置
with st.sidebar:
st.header("配置")
# 模板选择/创建
settings_path = Path("metagpt/ext/spo/settings")
existing_templates = [f.stem for f in settings_path.glob("*.yaml")]
template_mode = st.radio("模板模式", ["使用现有", "创建新模板"])
existing_templates = get_all_templates()
if template_mode == "使用现有":
template_name = st.selectbox("选择模板", existing_templates)
is_new_template = False
else:
template_name = st.text_input("新模板名称")
is_new_template = True
# LLM 设置
st.subheader("LLM 设置")
base_url = st.text_input("基础 URL", value="https://api.example.com")
api_key = st.text_input("API 密钥", type="password")
opt_model = st.selectbox(
"优化模型", ["gpt-4o-mini", "gpt-4o", "deepseek-chat", "claude-3-5-sonnet-20240620"], index=0
)
opt_temp = st.slider("优化温度", 0.0, 1.0, 0.7)
eval_model = st.selectbox(
"评估模型", ["gpt-4o-mini", "gpt-4o", "deepseek-chat", "claude-3-5-sonnet-20240620"], index=0
)
eval_temp = st.slider("评估温度", 0.0, 1.0, 0.3)
exec_model = st.selectbox(
"执行模型", ["gpt-4o-mini", "gpt-4o", "deepseek-chat", "claude-3-5-sonnet-20240620"], index=0
)
exec_temp = st.slider("执行温度", 0.0, 1.0, 0.0)
# 优化器设置
st.subheader("优化器设置")
initial_round = st.number_input("初始轮次", 1, 100, 1)
max_rounds = st.number_input("最大轮次", 1, 100, 10)
# 主要内容区域
st.header("模板配置")
if template_name:
template_real_name = get_template_path(template_name, is_new_template)
settings_path = Path("metagpt/ext/spo/settings")
template_path = settings_path / template_real_name
template_data = load_yaml_template(template_path)
if "current_template" not in st.session_state or st.session_state.current_template != template_name:
st.session_state.current_template = template_name
st.session_state.qas = template_data.get("qa", [])
# 编辑模板部分
prompt = st.text_area("提示词", template_data.get("prompt", ""), height=100)
requirements = st.text_area("要求", template_data.get("requirements", ""), height=100)
# 问答部分
st.subheader("问答示例")
# 添加新问答按钮
if st.button("添加新问答"):
st.session_state.qas.append({"question": "", "answer": ""})
# 编辑问答
new_qas = []
for i in range(len(st.session_state.qas)):
st.markdown(f"**问答 #{i + 1}**")
col1, col2, col3 = st.columns([45, 45, 10])
with col1:
question = st.text_area(
f"问题 {i + 1}", st.session_state.qas[i].get("question", ""), key=f"q_{i}", height=100
)
with col2:
answer = st.text_area(
f"答案 {i + 1}", st.session_state.qas[i].get("answer", ""), key=f"a_{i}", height=100
)
with col3:
if st.button("🗑️", key=f"delete_{i}"):
st.session_state.qas.pop(i)
st.rerun()
new_qas.append({"question": question, "answer": answer})
# 保存模板按钮
if st.button("保存模板"):
new_template_data = {"prompt": prompt, "requirements": requirements, "count": None, "qa": new_qas}
save_yaml_template(template_path, new_template_data, is_new_template)
st.session_state.qas = new_qas
st.success(f"模板已保存到 {template_path}")
st.subheader("当前模板预览")
preview_data = {"qa": new_qas, "requirements": requirements, "prompt": prompt}
st.code(yaml.dump(preview_data, allow_unicode=True), language="yaml")
st.subheader("优化日志")
log_container = st.empty()
class StreamlitSink:
def write(self, message):
current_logs = st.session_state.get("logs", [])
current_logs.append(message.strip())
st.session_state.logs = current_logs
log_container.code("\n".join(current_logs), language="plaintext")
streamlit_sink = StreamlitSink()
_logger.remove()
def prompt_optimizer_filter(record):
return "optimizer" in record["name"].lower()
_logger.add(
streamlit_sink.write,
format="{time:YYYY-MM-DD HH:mm:ss.SSS} | {level: <8} | {name}:{function}:{line} - {message}",
filter=prompt_optimizer_filter,
)
_logger.add(METAGPT_ROOT / "logs/{time:YYYYMMDD}.txt", level="DEBUG")
# 开始优化按钮
if st.button("开始优化"):
try:
# Initialize LLM
SPO_LLM.initialize(
optimize_kwargs={"model": opt_model, "temperature": opt_temp, "base_url": base_url,
"api_key": api_key},
evaluate_kwargs={"model": eval_model, "temperature": eval_temp, "base_url": base_url,
"api_key": api_key},
execute_kwargs={"model": exec_model, "temperature": exec_temp, "base_url": base_url,
"api_key": api_key},
)
# Create optimizer instance
optimizer = PromptOptimizer(
optimized_path=str(workspace_dir),
initial_round=initial_round,
max_rounds=max_rounds,
template=f"{template_real_name}",
name=template_name,
)
# Run optimization with progress bar
with st.spinner("Optimizing prompts..."):
optimizer.optimize()
st.success("优化完成!")
st.header("优化结果")
prompt_path = optimizer.root_path / "prompts"
result_data = optimizer.data_utils.load_results(prompt_path)
st.session_state.optimization_results = result_data
except Exception as e:
st.error(f"发生错误:{str(e)}")
_logger.error(f"优化过程中出错:{str(e)}")
if st.session_state.optimization_results:
st.header("优化结果")
display_optimization_results(st.session_state.optimization_results)
st.markdown("---")
st.subheader("测试优化后的提示词")
col1, col2 = st.columns(2)
with col1:
test_prompt = st.text_area("优化后的提示词", value="", height=200, key="test_prompt")
with col2:
test_question = st.text_area("你的问题", value="", height=200, key="test_question")
if st.button("测试提示词"):
if test_prompt and test_question:
try:
with st.spinner("正在生成回答..."):
SPO_LLM.initialize(
optimize_kwargs={"model": opt_model, "temperature": opt_temp, "base_url": base_url,
"api_key": api_key},
evaluate_kwargs={"model": eval_model, "temperature": eval_temp, "base_url": base_url,
"api_key": api_key},
execute_kwargs={"model": exec_model, "temperature": exec_temp, "base_url": base_url,
"api_key": api_key},
)
llm = SPO_LLM.get_instance()
messages = [{"role": "user", "content": f"{test_prompt}\n\n{test_question}"}]
async def get_response():
return await llm.responser(request_type=RequestType.EXECUTE, messages=messages)
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
try:
response = loop.run_until_complete(get_response())
finally:
loop.close()
st.subheader("回答:")
st.markdown(response)
except Exception as e:
st.error(f"生成回答时出错:{str(e)}")
else:
st.warning("请输入提示词和问题。")
if __name__ == "__main__":
main()