LLM Samplers Documentation

A Python library for advanced LLM sampling techniques, providing a collection of sophisticated sampling methods for language models.

This documentation is also available online at llm-samplers.readthedocs.io.

Contents:

• Installation
  • From PyPI
  • From Source
  • Requirements
• Usage
  • Basic Usage
  • Using with Custom PyTorch Models
  • Using with Other PyTorch Models
  • Model Compatibility
  • Available Samplers
• API Reference
  • Base Sampler
  • Beam Search Sampler
  • Temperature Sampler
  • Top-K Sampler
  • Top-P (Nucleus) Sampler
  • Min-P Sampler
  • Anti-Slop Sampler
  • XTC (Exclude Top Choices) Sampler
  • QAlign Sampler
• Sampling Methods
  • Temperature Sampling
  • Top-K Sampling
  • Top-P (Nucleus) Sampling
  • Min-P Sampling
  • Anti-Slop Sampling
  • XTC (Exclude Top Choices) Sampling
  • QAlign Sampling
  • Beam Search
• Examples
  • Basic Text Generation
  • Comparing Different Samplers
  • Customizing Generation Parameters
  • Using Anti-Slop for Higher Quality
  • Using Beam Search for Deterministic Outputs
• Contributing
  • Development Setup
  • Code Style
  • Writing Tests
  • Pull Request Process
  • Creating a New Sampler
  • Documentation
• License

Features

• Temperature Scaling

• Top-K Sampling

• Top-P (Nucleus) Sampling

• Min-P Sampling

• Anti-Slop Sampling

• XTC (Exclude Top Choices) Sampling

• Beam Search

• QAlign (MCMC Test-Time Alignment) Sampling

Indices and tables

• Index

• Module Index

• Search Page