sdar4b-trace-sft-esft-intent / train_results.json
autoprogrammer's picture
SDAR-4B trace_sft on ESFT-intent (final)
c20ea2f verified
{
"effective_tokens_per_sec": 151.98800620935165,
"epoch": 3.0,
"total_flos": 1.0795094318314947e+18,
"train_loss": 0.14432617937002265,
"train_runtime": 3312.6808,
"train_samples_per_second": 13.186,
"train_steps_per_second": 0.206
}