quasar-p1 / train_state.json
michael-chan-000's picture
Upload model
3b45902 verified
Raw
History Blame Contribute Delete
284 Bytes
{
"global_step": 5500,
"early_stopped": true,
"early_stop_reason": "patience exhausted (5/5 evals without \u22650.10% relative improvement, smoothed=0.3145, best=0.3132 @ step 5000, direction=lower)",
"early_stop_best_step": 5000,
"max_steps": 8000,
"phase": "sft_mixed"
}