AMOR-Mamba2-440M / training_summary.json
FlyinGodzilla's picture
Upload canonical AMOR checkpoint with provenance
def39ed verified
Raw History Blame Contribute Delete
551 Bytes
{
"model": "amor",
"scale": "440m",
"final_val_loss": 2.9757252883911134,
"final_val_ppl": 19.60383652966566,
"best_val_loss": 2.9758635330200196,
"best_val_ppl": 19.606546842110593,
"total_steps": 18255,
"tokens_trained": 8973017088,
"training_hours": 14.797530925538805,
"n_params": 448650496,
"flops_per_token": 1634942976.0,
"world_size": 1,
"linear_flops": 778505772072960,
"attention_flops": 18916646584320,
"ssm_flops": 6184752906240,
"final_fire_rate": 0.4057179203764526,
"final_alpha": 0.7287433220977273
}