Backtrack Sampler is a framework designed for experimenting with custom sampling strategies for language models (LLMs), enabling the ability to rewind and revise generated tokens. It allows developers to create and test their own token generation strategies by providing a base structure for manipulating logits and probabilities, making it a flexible tool for those interested in fine-tuning the behavior of LLMs.
Features
- Customizable token generation strategies
- Ability to backtrack and revise generated tokens
- Integration with models from Transformers and Llama.cpp
- Anti-slop strategy to prevent undesirable token generation
- Creative writing strategy to enhance model creativity by modifying token selection
- Easy-to-implement strategies using the base strategy class
Categories
FrameworksLicense
MIT LicenseFollow Backtrack Sampler
Other Useful Business Software
Full-stack observability with actually useful AI | Grafana Cloud
Built on open standards like Prometheus and OpenTelemetry, Grafana Cloud includes Kubernetes Monitoring, Application Observability, Incident Response, plus the AI-powered Grafana Assistant. Get started with our generous free tier today.
Rate This Project
Login To Rate This Project
User Reviews
Be the first to post a review of Backtrack Sampler!