LLM Finetuning Toolkit
A config-based command-line tool for running and comparing LLM fine-tuning experiments, controlling prompts, models, optimization strategy and testing from a single YAML file.
LLM Finetuning Toolkit is a CLI tool for launching a series of LLM fine-tuning experiments from one YAML configuration file, covering data ingestion, model definition, training, inference and quality-assurance testing in a single pipeline. A generated starter config file gives a base for basic runs, and the same file format scales up to experiments that sweep across multiple prompt templates, models and optimization techniques.
It installs with pipx or pip as the llm-toolkit package and exposes an llmtune command to generate a config and run an experiment. Optional Flash Attention 2 support can be enabled through the config once the flash-attn package is installed.