129 lines
5.5 KiB
Python
129 lines
5.5 KiB
Python
|
import gradio as gr
|
|||
|
|
import pandas as pd
|
|||
|
|
# import tempfile
|
|||
|
|
from http import HTTPStatus
|
|||
|
|
import dashscope
|
|||
|
|
from dashscope import Generation
|
|||
|
|
import os
|
|||
|
|
# import logging
|
|||
|
|
from testAny import check_df_english, check_df_tags
|
|||
|
|
import concurrent.futures
|
|||
|
|
import datetime
|
|||
|
|
|
|||
|
|
# 设置日志记录
|
|||
|
|
# logging.basicConfig(level=logging.INFO)
|
|||
|
|
|
|||
|
|
dashscope.api_key = os.getenv("DASHSCOPE_API_KEY") # Vincent's API key
|
|||
|
|
|
|||
|
|
def response(prompt, instruction=None):
|
|||
|
|
messages = [{'role': 'user', 'content': prompt}]
|
|||
|
|
if instruction is not None: # 如果提供了指令，则添加到messages中
|
|||
|
|
messages.insert(0, {'role': 'system', 'content': instruction})
|
|||
|
|
|
|||
|
|
try:
|
|||
|
|
response = Generation.call(model='qwen-plus',
|
|||
|
|
messages=messages,
|
|||
|
|
seed=1234,
|
|||
|
|
result_format='message',
|
|||
|
|
stream=False,
|
|||
|
|
incremental_output=False,
|
|||
|
|
temperature=0.85,
|
|||
|
|
top_p=0.9,
|
|||
|
|
top_k=999
|
|||
|
|
)
|
|||
|
|
if response.status_code == HTTPStatus.OK:
|
|||
|
|
message = response.output.choices[0]['message']['content']
|
|||
|
|
return message
|
|||
|
|
else:
|
|||
|
|
print('Request id: %s, Status code: %s, error code: %s, error message: %s' % (
|
|||
|
|
response.request_id, response.status_code,
|
|||
|
|
response.code, response.message
|
|||
|
|
))
|
|||
|
|
return f"Error: Could not generate response with Status code: {response.status_code}, error code: {response.code}"
|
|||
|
|
except Exception as e:
|
|||
|
|
print(f"Failed to generate response: {e}")
|
|||
|
|
return f"Error: Failed to generate response due to an error."
|
|||
|
|
|
|||
|
|
def format_full_prompt(df, introduction):
|
|||
|
|
# 为每个 row 创建 context，拼接RAG1和2
|
|||
|
|
df['context'] = df.apply(lambda row: f"{row['RAG1']}-{row['RAG2']}", axis=1)
|
|||
|
|
|
|||
|
|
# 准备用于 format 的字典
|
|||
|
|
column_list = df.drop('full_prompt', axis=1).columns.tolist() # 去除full_prompt列，其他的都为参数
|
|||
|
|
format_dict = df[column_list].apply(lambda x: dict(zip(x.index, x)), axis=1)
|
|||
|
|
if len(introduction) >= 200:
|
|||
|
|
df['full_prompt'] = introduction
|
|||
|
|
# 使用 apply() 和 lambda 函数格式化 full_prompt 列
|
|||
|
|
df['full_prompt'] = df.apply(lambda row: row['full_prompt'].format(**format_dict[row.name]), axis=1)
|
|||
|
|
|
|||
|
|
# 删除临时创建的 context 列
|
|||
|
|
df.drop(columns=['context'], inplace=True)
|
|||
|
|
return df
|
|||
|
|
|
|||
|
|
def process_xlsx(xlsx_file, instruction=None, loops=1):
|
|||
|
|
try:
|
|||
|
|
# 读取xlsx文件到pandas DataFrame
|
|||
|
|
df = pd.read_excel(xlsx_file)
|
|||
|
|
# 格式化prompts
|
|||
|
|
formatted_df = format_full_prompt(df, instruction)
|
|||
|
|
if loops > 1:
|
|||
|
|
df_list = [formatted_df.copy() for _ in range(loops - 1)]
|
|||
|
|
# 使用pd.concat一次性合并所有副本
|
|||
|
|
formatted_df = pd.concat([formatted_df] + df_list, ignore_index=True)
|
|||
|
|
|
|||
|
|
# 调用response时，根据instruction是否为None自动处理
|
|||
|
|
formatted_df['Response'] = formatted_df['full_prompt'].apply(lambda prompt: response(prompt, instruction))
|
|||
|
|
|
|||
|
|
# check df with tags and english
|
|||
|
|
formatted_df = check_df_tags(formatted_df)
|
|||
|
|
formatted_df = check_df_english(formatted_df)
|
|||
|
|
|
|||
|
|
# 使用一个文件路径保存处理后的xlsx
|
|||
|
|
date_str = datetime.datetime.now().strftime("%m%d%H")
|
|||
|
|
times_str = str(20 * loops)
|
|||
|
|
output_path = 'output'
|
|||
|
|
if not os.path.exists(output_path):
|
|||
|
|
os.makedirs(output_path)
|
|||
|
|
file_name = f"{date_str}_{times_str}times_output.xlsx"
|
|||
|
|
file_path = os.path.join(output_path, file_name)
|
|||
|
|
formatted_df.to_excel(file_path, index=False, engine='openpyxl')
|
|||
|
|
return formatted_df, file_path
|
|||
|
|
except Exception as e:
|
|||
|
|
print(f"Failed to process xlsx: {e}")
|
|||
|
|
return None, None
|
|||
|
|
|
|||
|
|
def main():
|
|||
|
|
with gr.Blocks() as demo:
|
|||
|
|
gr.Markdown("### 大模型xlsx处理工具")
|
|||
|
|
with gr.Accordion("输入说明"):
|
|||
|
|
gr.Markdown("请上传一个xlsx文件，文件应包含prompts。")
|
|||
|
|
system_instruction = gr.Textbox(label="System Instruction", lines=2,
|
|||
|
|
value=" ")
|
|||
|
|
slider = gr.Slider(minimum=1, maximum=10, step=1, label="循环次数", value=1)
|
|||
|
|
|
|||
|
|
file_input = gr.File(label="上传xlsx文件")
|
|||
|
|
submit_button = gr.Button("处理xlsx")
|
|||
|
|
|
|||
|
|
output_table = gr.Dataframe(label="处理后的数据")
|
|||
|
|
output_file = gr.File(label="下载处理后的文件")
|
|||
|
|
clear_data = gr.ClearButton(components=[output_table, output_file], value="Clear processed data")
|
|||
|
|
clear_all = gr.ClearButton(components=[file_input, output_table, output_file], value="Clear console")
|
|||
|
|
def update_output(xlsx_file, instruction, loops):
|
|||
|
|
if xlsx_file is not None:
|
|||
|
|
formatted_df, tmp_path = process_xlsx(xlsx_file, instruction, loops)
|
|||
|
|
# 使用线程池并行处理每一行
|
|||
|
|
with concurrent.futures.ThreadPoolExecutor() as executor:
|
|||
|
|
formatted_df['Response'] = list(executor.map(response, formatted_df['full_prompt'], [instruction] * len(formatted_df)))
|
|||
|
|
|
|||
|
|
if formatted_df is not None:
|
|||
|
|
return formatted_df, tmp_path # 返回DataFrame和文件路径
|
|||
|
|
|
|||
|
|
submit_button.click(fn=update_output, inputs=[file_input, system_instruction, slider],
|
|||
|
|
outputs=[output_table, output_file])
|
|||
|
|
|
|||
|
|
demo.launch()
|
|||
|
|
|
|||
|
|
|
|||
|
|
if __name__ == "__main__":
|
|||
|
|
main()
|