bgc-accession / checkpoint-4000 /trainer_state.json
root
First version of the bgc-accession model and tokenizer.
3f66fdf
raw
history blame
1.34 kB
{
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 22.22222222222222,
"global_step": 4000,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 2.78,
"learning_rate": 3.62962962962963e-05,
"loss": 0.0119,
"step": 500
},
{
"epoch": 5.56,
"learning_rate": 3.259259259259259e-05,
"loss": 0.0015,
"step": 1000
},
{
"epoch": 8.33,
"learning_rate": 2.888888888888889e-05,
"loss": 0.0007,
"step": 1500
},
{
"epoch": 11.11,
"learning_rate": 2.5185185185185187e-05,
"loss": 0.0003,
"step": 2000
},
{
"epoch": 13.89,
"learning_rate": 2.1481481481481485e-05,
"loss": 0.0002,
"step": 2500
},
{
"epoch": 16.67,
"learning_rate": 1.7777777777777777e-05,
"loss": 0.0001,
"step": 3000
},
{
"epoch": 19.44,
"learning_rate": 1.4074074074074075e-05,
"loss": 0.0001,
"step": 3500
},
{
"epoch": 22.22,
"learning_rate": 1.037037037037037e-05,
"loss": 0.0001,
"step": 4000
}
],
"max_steps": 5400,
"num_train_epochs": 30,
"total_flos": 8325643182451200.0,
"trial_name": null,
"trial_params": null
}