NanoMath-R / training_summary.json
agmadaasa's picture
Initial NanoMath-R release
084b790 verified
Raw
History Blame Contribute Delete
447 Bytes
{
"accelerator_hours": 23.124297935119447,
"block_applications_processed": 12582912000,
"estimated_training_flops": 601157982486528000,
"optimizer_steps_completed": 6000,
"peak_accelerator_memory_bytes": 2956411904,
"step": 5999,
"stored_parameters": 103834368,
"tokens_processed": 393216000,
"training_wall_seconds": 41623.736283215,
"validation_loss": 0.12996602058410645,
"validation_loss_step": 5750,
"world_size": 2
}