Skip to content
New issue

Add fine-tuning support for text-generation

Past due by over 1 year 56% complete

Is your feature request related to a problem? Please describe.

Currently we only support Prompt tuning based methods for tuning base models. These work great for larger language models, however, lacks a bit for smaller models in terms of quality. To enable more flexibility and choice for training, we want to enable fine-tuning of certain transformer based…

Is your feature request related to a problem? Please describe.

Currently we only support Prompt tuning based methods for tuning base models. These work great for larger language models, however, lacks a bit for smaller models in terms of quality. To enable more flexibility and choice for training, we want to enable fine-tuning of certain transformer based models using HF Trainer: https://huggingface.co/docs/transformers/main_classes/trainer

Specifics:

  1. Caikit Task: Text-generation
  2. Eval task: Summarization
  3. Test Models: TBD
  4. Test Datasets: TBD
  5. GPU Support: Single Node, multi-GPU support
  6. Training stream data model: https://github.com/caikit/caikit-nlp/blob/main/caikit_nlp/data_model/generation.py#L54
  7. Data formats (for caikit-runtime): csv

https://huggingface.co/docs/transformers/main_classes/trainer

Loading