View and submit LLM evaluations
Convert Hugging Face model repo to Safetensors
Evaluate open LLMs in the languages of LATAM and Spain.
Retrain models for new data at edge devices
Upload a machine learning model to Hugging Face Hub
Convert and upload model files for Stable Diffusion
Measure over-refusal in LLMs using OR-Bench
Browse and submit LLM evaluations
Measure BERT model performance using WASM and WebGPU
Analyze model errors with interactive pages
Create and manage ML pipelines with ZenML Dashboard
Evaluate RAG systems with visual analytics
Push a ML model to Hugging Face Hub
Hallucinations Leaderboard is a tool designed for evaluating and benchmarking large language models (LLMs). It provides a platform to view and submit evaluations of model performance, with a focus on understanding and mitigating hallucinations—instances where models produce inaccurate or non-factual information.
• Leaderboard System: Compare performance of different LLMs based on hallucination metrics. • Benchmarking Tools: Access standardized tests and evaluations for assessing model accuracy. • Customizable Metrics: Define and apply specific criteria for measuring hallucinations. • Model Comparison: Directly compare multiple models side-by-side. • Submission Interface: Easily submit your own evaluations for inclusion in the leaderboard. • Filtering and Sorting: Narrow down results by model size, architecture, or performance thresholds. • Real-Time Updates: Stay current with the latest evaluations and benchmarks.
What is the purpose of Hallucinations Leaderboard?
The purpose is to provide a centralized platform for evaluating and comparing LLMs, with a focus on reducing hallucinations and improving model accuracy.
How do I submit my own evaluations?
To submit evaluations, use the submission interface on the platform. Ensure your results align with the defined metrics and criteria.
Why is tracking hallucinations important?
Hallucinations can lead to misinformation. Tracking them helps improve model reliability and trustworthiness in real-world applications.