|
|
|
|
|
""" |
|
Gradio UI – v2.1 (Leaderboard · Data Viewer · Prompt-to-Leaderboard) |
|
""" |
|
|
|
from __future__ import annotations |
|
from pathlib import Path |
|
import gradio as gr |
|
|
|
|
|
from tabs.leaderboard_tab import create_leaderboard_tab |
|
from tabs.data_viewer_tab import create_data_viewer_tab |
|
from tabs.data_viewer_side_by_side_tab import create_data_viewer_side_by_side_tab |
|
|
|
|
|
|
|
|
|
|
|
with gr.Blocks(title="DeepResearch Bench") as demo: |
|
|
|
|
|
gr.HTML(""" |
|
<style> |
|
.title-block{ |
|
/* 渐变文字效果 - 改进版 */ |
|
background: linear-gradient(to right, #009CFF, #823AFF); |
|
background: -webkit-linear-gradient(to right, #009CFF, #823AFF); |
|
background: -moz-linear-gradient(to right, #009CFF, #823AFF); |
|
-webkit-background-clip: text; |
|
-webkit-text-fill-color: transparent; |
|
background-clip: text; |
|
color: transparent; |
|
|
|
text-align: center; |
|
font-size: 2rem; |
|
font-weight: 700; |
|
margin: 0 0 1rem 0; |
|
padding-bottom: 0.2rem; |
|
display: inline-block; /* 重要:确保渐变效果正常 */ |
|
width: 100%; /* 确保居中对齐 */ |
|
} |
|
.intro-block{ |
|
text-align:center; |
|
margin-bottom:1.25rem; |
|
line-height:2; |
|
} |
|
.intro-block a{ |
|
color:#0a58ca; |
|
text-decoration:none; |
|
margin:0 .3rem; |
|
} |
|
.intro-block a:hover{ text-decoration:underline; } |
|
</style> |
|
""") |
|
|
|
|
|
gr.HTML(""" |
|
<div class="title-block"> |
|
DeepResearch Bench: A Comprehensive Benchmark for Deep Research Agents |
|
</div> |
|
|
|
<div class="intro-block"> |
|
The research aims to comprehensively evaluate the capabilities of Deep Research Agents.<br> |
|
<a href="#" target="_blank">Code</a> | |
|
<a href="https://deepresearch-bench.github.io" target="_blank">Website</a> | |
|
<a href="#" target="_blank">Paper</a> | |
|
<a href="#" target="_blank">Eval Dataset</a> | |
|
Total models: 16 | Last Update: 28 May 2025 |
|
</div> |
|
""") |
|
|
|
|
|
with gr.Tabs(): |
|
create_leaderboard_tab() |
|
create_data_viewer_side_by_side_tab() |
|
create_data_viewer_tab() |
|
|
|
with gr.Tab("💬Prompt-to-Leaderboard"): |
|
gr.Markdown( |
|
""" |
|
🚧 **Prompt-to-Leaderboard** module not implemented yet. |
|
Planned: inspect how individual prompts affect overall model ranking. |
|
""" |
|
) |
|
|
|
|
|
|
|
|
|
if __name__ == "__main__": |
|
demo.launch() |
|
|