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
385 B
JSON
13 lines
385 B
JSON
{
|
|
"epoch": 1.0,
|
|
"eval_loss": 0.87261962890625,
|
|
"eval_runtime": 84.2749,
|
|
"eval_samples_per_second": 11.866,
|
|
"eval_steps_per_second": 11.866,
|
|
"num_input_tokens_seen": 62891360,
|
|
"total_flos": 2.69902163257983e+18,
|
|
"train_loss": 0.942430813171684,
|
|
"train_runtime": 79344.5851,
|
|
"train_samples_per_second": 3.768,
|
|
"train_steps_per_second": 0.236
|
|
} |