Supastrikas-004/evaluation-framework

Agentic Evaluation Framework — Hugging Face Space

This Gradio app evaluates and compares many AI agents across tasks (QA, summarization, reasoning...) using lightweight scorers and visualizations.

How to use

1. 1.Upload a CSV/JSON/JSONL file with columns: prompt, response, task, agent, reference (reference optional).
2. 2.Click Run Evaluation.
3. 3.View per-task spider charts, heatmaps, bar plots in the Gallery, inspect per-example metrics in the table, and download the CSV report.

If no file is uploaded, a small synthetic demo dataset will be evaluated.

Deploying

• —Push this repo into a Hugging Face Space (Gradio). The requirements.txt will be installed automatically.

Notes & Limitations

• —Models used are lightweight but still require CPU memory (no Java).
• —If reference is missing, hallucination/accuracy signals will be reduced.
• —Coherence metric is a placeholder heuristic — you can replace it with grammar/perplexity models if desired.