Turn preference feedback into a concrete training config.
What it does
Selects a preference-optimization method and supplies a concrete training configuration for paired preferences or unpaired feedback.
When to use it
Use when preference pairs or thumbs-up/down feedback exist, when choosing a preference-optimization method, or when debugging or configuring DPO.
How to use it
Give it preference pairs or unpaired feedback routed to preference optimization; it returns a validated method choice and concrete configuration values.
What you provide
- Your files
Latest review
What you need · 1
The supplied training code uses TRL configuration classes and trainers, with processing_class rather than tokenizer as the trainer argument.
Access · 0
No additional actions listed.