Best Model Save in Ludwig

[msmmpts]

Hi,

I was running a fine-tuning exercise for Llama 2 model for 10 epochs. Logs however indicated that the best model epoch was 7.

Does the adapter_config.json and adapter_model.bin correspond to epoch 7(best epoch) that was reported as the best?

Thanks

[tgaddair]

Hey @msmmpts, yes, the final adapter weights that get saved as model_weights are taken from the best checkpoint (in this case, epoch 7).

[msmmpts]

Thanks