Jakub
67061cf373
Document six completed LR runs as preparation for full Aya training on Perun. Record selection of LR 0.00010 by lowest final validation loss, selected-model lm-eval and partial SK-bench results, reproducible provenance, and dataset-access/context-truncation limitations. Preserve existing 30k and Titan material.
13 lines
388 B
JSON
13 lines
388 B
JSON
{
|
|
"epoch": 1.0,
|
|
"eval_loss": 0.8976690173149109,
|
|
"eval_runtime": 82.9721,
|
|
"eval_samples_per_second": 12.052,
|
|
"eval_steps_per_second": 12.052,
|
|
"num_input_tokens_seen": 62891360,
|
|
"total_flos": 2.69902163257983e+18,
|
|
"train_loss": 0.9813241457490072,
|
|
"train_runtime": 79710.3252,
|
|
"train_samples_per_second": 3.751,
|
|
"train_steps_per_second": 0.234
|
|
} |