Update app.py
Browse files
app.py
CHANGED
|
@@ -1,82 +1,147 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import requests
|
| 3 |
import json
|
|
|
|
|
|
|
| 4 |
|
| 5 |
# ----------------- 配置你的服务器API地址 -----------------
|
| 6 |
# 将IP地址和端口替换为你自己的
|
| 7 |
SERVER_URL = "http://103.235.229.133:7000/predict"
|
| 8 |
# -------------------------------------------------------
|
| 9 |
|
| 10 |
-
#
|
| 11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
-
def call_api(
|
| 14 |
"""
|
| 15 |
-
这个函数会被Gradio
|
|
|
|
| 16 |
"""
|
| 17 |
-
|
| 18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
| 20 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
payload = {
|
| 22 |
-
"model_name":
|
| 23 |
"data_csv": csv_data,
|
| 24 |
-
"only_final_result":
|
| 25 |
}
|
| 26 |
|
| 27 |
try:
|
| 28 |
-
# 发送POST请求
|
| 29 |
-
|
|
|
|
|
|
|
| 30 |
|
| 31 |
-
#
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
return pretty_json, dataframe_result
|
| 43 |
else:
|
| 44 |
-
#
|
| 45 |
-
|
| 46 |
-
|
| 47 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
except requests.exceptions.RequestException as e:
|
| 49 |
-
#
|
| 50 |
-
|
| 51 |
-
|
|
|
|
|
|
|
|
|
|
| 52 |
|
| 53 |
-
# 使用Gradio Blocks
|
| 54 |
-
with gr.Blocks(theme=gr.themes.
|
| 55 |
-
gr.Markdown(
|
| 56 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
|
| 58 |
with gr.Row():
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
)
|
| 66 |
-
|
|
|
|
| 67 |
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
gr.
|
| 72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
|
|
|
|
| 74 |
submit_btn.click(
|
| 75 |
fn=call_api,
|
| 76 |
-
inputs=
|
| 77 |
outputs=[output_json, output_dataframe]
|
| 78 |
)
|
| 79 |
|
| 80 |
# 启动Gradio应用
|
| 81 |
if __name__ == "__main__":
|
|
|
|
| 82 |
demo.launch()
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import requests
|
| 3 |
import json
|
| 4 |
+
import pandas as pd
|
| 5 |
+
import io
|
| 6 |
|
| 7 |
# ----------------- 配置你的服务器API地址 -----------------
|
| 8 |
# 将IP地址和端口替换为你自己的
|
| 9 |
SERVER_URL = "http://103.235.229.133:7000/predict"
|
| 10 |
# -------------------------------------------------------
|
| 11 |
|
| 12 |
+
# 默认的模型名称和输入数据,方便用户测试
|
| 13 |
+
DEFAULT_MODEL_NAME = "/data/tsq/MODELS/Qwen2.5-1.5B-Instruct"
|
| 14 |
+
DEFAULT_DATAFRAME_VALUE = [
|
| 15 |
+
["France", "capital city of", "Paris", "she traveled to France for the first time"],
|
| 16 |
+
["Germany", "capital city of", "Berlin", "the capital of Germany is Berlin"]
|
| 17 |
+
]
|
| 18 |
|
| 19 |
+
def call_api(model_name, only_final_result, data_df):
|
| 20 |
"""
|
| 21 |
+
这个函数会被Gradio调用,它负责收集UI上的输入,
|
| 22 |
+
将其转换成API需要的格式,然后发送请求。
|
| 23 |
"""
|
| 24 |
+
# 1. 数据校验
|
| 25 |
+
if not model_name.strip():
|
| 26 |
+
# 使用Gradio的错误提示功能
|
| 27 |
+
raise gr.Error("模型名称不能为空!")
|
| 28 |
+
if data_df is None or data_df.empty:
|
| 29 |
+
raise gr.Error("输入数据不能为空!请在表格中至少添加一行数据。")
|
| 30 |
|
| 31 |
+
# 2. 将输入的DataFrame转换成API需要的CSV字符串
|
| 32 |
+
# API需要 'id' 列,我们在这里自动生成
|
| 33 |
+
data_df_with_id = data_df.copy()
|
| 34 |
+
# 为id列生成唯一的、格式化的ID,例如 000, 001, 002...
|
| 35 |
+
data_df_with_id.insert(0, 'id', [f'{i:03d}' for i in range(len(data_df_with_id))])
|
| 36 |
+
|
| 37 |
+
# 将DataFrame转换为CSV字符串
|
| 38 |
+
# 使用io.StringIO来在内存中处理字符串,避免写入文件
|
| 39 |
+
output = io.StringIO()
|
| 40 |
+
data_df_with_id.to_csv(output, index=False)
|
| 41 |
+
csv_data = output.getvalue()
|
| 42 |
+
|
| 43 |
+
# 3. 构建发送到API的JSON数据体
|
| 44 |
payload = {
|
| 45 |
+
"model_name": model_name,
|
| 46 |
"data_csv": csv_data,
|
| 47 |
+
"only_final_result": only_final_result
|
| 48 |
}
|
| 49 |
|
| 50 |
try:
|
| 51 |
+
# 4. 发送POST请求
|
| 52 |
+
# 添加流式传输和更长的超时时间以应对长时间运行的任务
|
| 53 |
+
response = requests.post(SERVER_URL, json=payload, timeout=300, stream=True)
|
| 54 |
+
response.raise_for_status() # 如果状态码不是2xx,则抛出HTTPError
|
| 55 |
|
| 56 |
+
# 解析返回的JSON数据
|
| 57 |
+
result_json = response.json()
|
| 58 |
+
|
| 59 |
+
# 5. 处理和返回结果
|
| 60 |
+
# 从结果中提取'result'键,它应该是一个字典列表
|
| 61 |
+
dataframe_result_data = result_json.get("result", [])
|
| 62 |
+
|
| 63 |
+
# 将字典列表转换为Pandas DataFrame,以便Gradio的DataFrame组件更好地显示
|
| 64 |
+
if isinstance(dataframe_result_data, list) and len(dataframe_result_data) > 0:
|
| 65 |
+
output_df = pd.DataFrame(dataframe_result_data)
|
|
|
|
|
|
|
| 66 |
else:
|
| 67 |
+
output_df = pd.DataFrame() # 如果没有结果,返回一个空的DataFrame
|
| 68 |
+
|
| 69 |
+
return result_json, output_df
|
| 70 |
|
| 71 |
+
except requests.exceptions.HTTPError as e:
|
| 72 |
+
# 更详细地捕获HTTP错误
|
| 73 |
+
error_details = f"API返回错误 (状态码: {e.response.status_code}):\n{e.response.text}"
|
| 74 |
+
raise gr.Error(f"请求失败: {error_details}")
|
| 75 |
except requests.exceptions.RequestException as e:
|
| 76 |
+
# 捕获网络连接等问题
|
| 77 |
+
raise gr.Error(f"连接服务器失败: {e}")
|
| 78 |
+
except Exception as e:
|
| 79 |
+
# 捕获其他未知错误
|
| 80 |
+
raise gr.Error(f"发生未知错误: {e}")
|
| 81 |
+
|
| 82 |
|
| 83 |
+
# --- 使用Gradio Blocks创建更灵活、更美观的界面 ---
|
| 84 |
+
with gr.Blocks(theme=gr.themes.Glass(), css=".gradio-container {max-width: 1280px !important; margin: auto;}") as demo:
|
| 85 |
+
gr.Markdown(
|
| 86 |
+
"""
|
| 87 |
+
# 🚀 Ml_patch 函数推理接口
|
| 88 |
+
一个为 `Ml_patch` 函数设计的交互式Web界面。请在下方配置模型参数并输入您要测试的数据。
|
| 89 |
+
"""
|
| 90 |
+
)
|
| 91 |
|
| 92 |
with gr.Row():
|
| 93 |
+
# 左侧列:输入配置
|
| 94 |
+
with gr.Column(scale=2):
|
| 95 |
+
gr.Markdown("### 1. 配置模型和参数")
|
| 96 |
+
with gr.Group():
|
| 97 |
+
model_name_input = gr.Textbox(
|
| 98 |
+
label="模型名称 (model_name)",
|
| 99 |
+
value=DEFAULT_MODEL_NAME,
|
| 100 |
+
info="请输入部署在服务器上的模型路径。"
|
| 101 |
+
)
|
| 102 |
+
only_final_result_input = gr.Checkbox(
|
| 103 |
+
label="仅返回最终结果 (only_final_result)",
|
| 104 |
+
value=False,
|
| 105 |
+
info="勾选此项,API将只返回最终聚合结果,而不是所有层的详细信息。"
|
| 106 |
+
)
|
| 107 |
+
|
| 108 |
+
gr.Markdown("### 2. 输入数据")
|
| 109 |
+
# 使用可编辑的DataFrame作为输入,用户可以动态增删行
|
| 110 |
+
data_input = gr.DataFrame(
|
| 111 |
+
label="数据表格",
|
| 112 |
+
headers=["subject", "relation", "object", "prompt_source"],
|
| 113 |
+
value=DEFAULT_DATAFRAME_VALUE,
|
| 114 |
+
row_count=(1, "dynamic"), # 至少1行,动态增加
|
| 115 |
+
col_count=(4, "fixed"),
|
| 116 |
+
interactive=True,
|
| 117 |
+
info="点击表格下方的 'Edit' (铅笔图标),然后点击 '+' 来添加新行,或点击垃圾桶图标删除行。"
|
| 118 |
)
|
| 119 |
+
|
| 120 |
+
submit_btn = gr.Button("提交推理", variant="primary", scale=1)
|
| 121 |
|
| 122 |
+
# 右侧列:输出结果
|
| 123 |
+
with gr.Column(scale=3):
|
| 124 |
+
gr.Markdown("### 3. 查看推理结果")
|
| 125 |
+
with gr.Tabs():
|
| 126 |
+
with gr.TabItem("结果表格"):
|
| 127 |
+
output_dataframe = gr.DataFrame(
|
| 128 |
+
label="推理结果详情",
|
| 129 |
+
interactive=False,
|
| 130 |
+
# 根据您的示例输出定义列,Gradio会自动匹配
|
| 131 |
+
headers=["subject", "prompt_source", "prompt_target", "layer_source", "is_correct_patched", "generations"]
|
| 132 |
+
)
|
| 133 |
+
with gr.TabItem("原始JSON响应"):
|
| 134 |
+
# 使用gr.JSON组件可以更好地格式化显示JSON
|
| 135 |
+
output_json = gr.JSON(label="API 返回的完整JSON")
|
| 136 |
|
| 137 |
+
# 定义按钮点击事件
|
| 138 |
submit_btn.click(
|
| 139 |
fn=call_api,
|
| 140 |
+
inputs=[model_name_input, only_final_result_input, data_input],
|
| 141 |
outputs=[output_json, output_dataframe]
|
| 142 |
)
|
| 143 |
|
| 144 |
# 启动Gradio应用
|
| 145 |
if __name__ == "__main__":
|
| 146 |
+
# 使用share=True可以生成一个公开链接,方便他人访问
|
| 147 |
demo.launch()
|