autotrain_ultra_low_final.yaml

project:
  name: exaone-nutrition-sft
  task: text-generation
  backend: transformers
data:
  train_path: d:\ExaoneFinetuning\data\final\train.csv
  valid_path: d:\ExaoneFinetuning\data\final\validation.csv
  input_column: text
  target_column: null
  format: sft-json-in-csv
model:
  hub_model: lg-electronics/exaone-3.5-7.8b-instruct
  max_length: 2048
  dtype: bfloat16
  quantization: bnb-4bit
training:
  epochs: 2
  batch_size: 2
  gradient_accumulation: 8
  lr: 0.0002
  warmup_ratio: 0.03
  weight_decay: 0.01
  save_steps: 200
  logging_steps: 20
  eval_steps: 200
  lora:
    r: 16
    alpha: 32
    dropout: 0.05
    target_modules: all-linear
output:
  dir: D:\ExaoneFinetuning\EXAONE_4_FINETUNING_NEW\outputs
  push_to_hub: false