Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,12 +1,33 @@
|
|
| 1 |
---
|
| 2 |
title: BioDesignBench Leaderboard
|
| 3 |
-
emoji:
|
| 4 |
-
colorFrom:
|
| 5 |
colorTo: purple
|
| 6 |
sdk: gradio
|
| 7 |
-
sdk_version:
|
| 8 |
app_file: app.py
|
| 9 |
pinned: false
|
|
|
|
| 10 |
---
|
| 11 |
|
| 12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
title: BioDesignBench Leaderboard
|
| 3 |
+
emoji: "\U0001F9EC"
|
| 4 |
+
colorFrom: blue
|
| 5 |
colorTo: purple
|
| 6 |
sdk: gradio
|
| 7 |
+
sdk_version: "4.44.0"
|
| 8 |
app_file: app.py
|
| 9 |
pinned: false
|
| 10 |
+
license: mit
|
| 11 |
---
|
| 12 |
|
| 13 |
+
# BioDesignBench Leaderboard
|
| 14 |
+
|
| 15 |
+
Evaluating LLM Agents on Protein Design via MCP Tools.
|
| 16 |
+
|
| 17 |
+
**Romero Lab, Duke University**
|
| 18 |
+
|
| 19 |
+
## Overview
|
| 20 |
+
|
| 21 |
+
BioDesignBench is the first comprehensive benchmark for evaluating LLM agents on
|
| 22 |
+
protein design tasks via MCP (Model Context Protocol) tool use. This leaderboard
|
| 23 |
+
tracks agent performance across 76 design tasks spanning 17 taxonomy cells
|
| 24 |
+
(5 DesignTaskTypes x 6 BiologicalContexts), scored on a 100-point rubric with
|
| 25 |
+
6 components: Approach, Orchestration, Quality, Feasibility, Novelty, Diversity.
|
| 26 |
+
|
| 27 |
+
## Features
|
| 28 |
+
|
| 29 |
+
- **Overall Leaderboard** — Mixed-ranking table with baselines and LLM agents
|
| 30 |
+
- **Taxonomy Breakdown** — Heatmap of per-cell scores across 17 taxonomy cells
|
| 31 |
+
- **Component Analysis** — Radar and bar charts comparing 6 scoring components
|
| 32 |
+
- **Benchmark vs User Mode** — Paired comparison of the same LLM in two modes
|
| 33 |
+
- **About** — Methodology, submission guide, and citation info
|