search

OptimalScale / LMFlow

An Extensible Toolkit for Finetuning and Inference of Large Foundation Models. Large Models for All.
https://optimalscale.github.io/LMFlow/
Apache License 2.0
8.11k stars 819 forks source link

Usability update #857

Closed wheresmyhair closed 2 weeks ago

wheresmyhair commented 2 weeks ago
  1. RM shell shebang fix
  2. Add padding side option, by default right (as what LMFlow did previously), when set to auto, will follow the tokenizer default padding_side
  3. Specify num_labels=1 for HFTextRegressionModel to avoid some rare bugs
  4. HFTextRegressionModel supports text_only and text2text data (only for inference, since training a reward model needs preference(paired) data)