Track, rank and evaluate open LLMs and chatbots
Display model benchmark results
Submit models for evaluation and view leaderboard
Evaluate LLM over-refusal rates with OR-Bench
Search for model performance across languages and benchmarks
Display leaderboard of language model evaluations
Convert PyTorch models to waifu2x-ios format
Browse and filter machine learning models by category and modality
Evaluate RAG systems with visual analytics
Multilingual Text Embedding Model Pruner
Convert a Stable Diffusion XL checkpoint to Diffusers and open a PR
Evaluate code generation with diverse feedback types
Submit deepfake detection models for evaluation
The Low-bit Quantized Open LLM Leaderboard is a platform designed to track, rank, and evaluate open large language models (LLMs) and chatbots with a focus on low-bit quantization. It provides insights into how these models perform when compressed to lower precision (e.g., 4-bit or 8-bit), enabling efficient deployment on edge devices. The leaderboard helps researchers and developers explore and compare the performance of various LLMs in resource-constrained environments.
• Model Benchmarking: Comprehensive evaluation of open-source LLMs using low-bit quantization.
• Quantization Tools: Built-in support for applying quantization techniques to reduce model size.
• Accuracy Metrics: Tracks performance across tasks like text generation, question answering, and conversational tasks.
• Efficiency Insights: Displays memory usage and inference speed for quantized models.
• Real-time Updates: Regularly updated leaderboard with the latest models and optimizations.
• Community Engagement: Open for contributions, fostering collaboration in the AI research community.
• Transparency: Detailed documentation of evaluation methodologies and metrics.
1. What models are included in the leaderboard?
The leaderboard includes a variety of open-source LLMs, focusing on models optimized for low-bit quantization. Popular models like BERT, GPT, and smaller variants are regularly featured.
2. How is the performance of quantized models measured?
Performance is measured using standard benchmarks like text generation quality, question answering accuracy, and inference speed. Additional metrics include memory usage and computational efficiency.
3. Can I use the leaderboard for commercial purposes?
Yes, the leaderboard is designed to support both research and practical applications. It provides valuable insights for deploying quantized models in real-world scenarios, such as edge devices.
4. How often is the leaderboard updated?
The leaderboard is updated regularly to include new models, improvements in quantization techniques, and feedback from the community.
5. Can I contribute to the leaderboard?
Absolutely! The platform encourages contributions, such as submitting new models, improving quantization techniques, or providing feedback on existing entries.