Large Language Models (LLMs)

Ludwig provides a declarative interface to support fine-tuning using a variety of Huggigngface and TorchText LLMs.

input_features:
- name: title
  type: text
  encoder:
    type: auto_transformer
    pretrained_model_name_or_path: bigscience/bloom-3b
    trainable: true
output_features:
- name: class
  type: category
trainer:
  learning_rate: 1.0e-05
  epochs: 3
backend:
  type: ray
  trainer:
    strategy: fsdp

Read the v0.7 blog post to learn about 50X optimizations.

See a demonstration using Ludwig Python API: