Files
rugpt3small_raw_baseline/results/training_metrics.json

16 lines
516 B
JSON
Raw Normal View History

{
"experiment": "raw_baseline",
"method": "raw",
"model_name": "ai-forever/rugpt3small_based_on_gpt2",
"base_dataset_repo": "Bykot/c4_ru_200k_split",
"train_docs": 198000,
"removed_docs": 0,
"eval_docs": 2000,
"eval_loss_before": 3.8492491245269775,
"perplexity_before": 46.95779053735424,
"eval_loss_after": 3.0527045726776123,
"perplexity_after": 21.172529702158773,
"train_runtime": 19125.8028,
"train_samples_per_second": 10.353,
"created_at_utc": "2026-05-05T08:11:53.641125+00:00"
}