Hugging Face's logo

Datasets:
SciVisAgentBench
/
SciVisAgentBench-tasks

Tasks:
Text-to-3D
Other
ArXiv:
License:
Dataset card Files
xet
 Community
2
SciVisAgentBench-tasks / eval_cases /molecular_vis /README.md
KuangshiAi
add two molecular vis workflow cases from Helgi Ingolfsson
40d081b
preview code
|
raw
history blame
1.33 kB

SciVisAgentBench: Molecular Visualization

This benchmark is designed to evaluate agents on molecular visualization tasks, specifically focusing on tools like VMD (Visual Molecular Dynamics).

Overview

The benchmark assesses agent capabilities across different levels of complexity:

  • Basic Actions (Easy): Simple, atomic commands and operations (Currently implemented).
  • Workflows (Medium): Sequences of actions forming a coherent pipeline (Planned).
  • Scientific Tasks (Hard): Complex, goal-oriented scientific analysis and visualization (Planned).

Setup

  1. Data Preparation:
    • Download the PDB file 1CRN from the RCSB PDB.
    • Place the downloaded file into the data/ folder in this directory.

Usage

You can run the evaluation cases defined in the YAML files (located in the actions/ directory) using either:

  • Promptfoo
  • The SciVisAgentBench Evaluation Framework

Refer to the main repository for detailed instructions on running the evaluation harness.

Future Extensions

SciVisAgentBench: MD Simulations

A placeholder for Molecular Dynamics (MD) simulation tasks is currently reserved. This will be expanded to include setting up, running, and analyzing simulations.