This is the default training configuration. We will leave it as is for the time being.
gradient_accumulation_steps: 4 micro_batch_size: 2 num_epochs: 4 optimizer: adamw_bnb_8bit lr_scheduler: cosine learning_rate: 0.0002
Last updated 10 months ago
