llama2-7b-TIM-fixemb / train_results.json
model-trial's picture
llama-7b model finetuned using TIM (fixemb) for trial
8502be8
raw
history blame contribute delete
315 Bytes
{
"epoch": 3.98,
"max_memory_allocated (GB)": 82.87,
"memory_allocated (GB)": 23.35,
"total_memory_available (GB)": 94.57,
"train_loss": 0.19149471794962883,
"train_runtime": 16857.8956,
"train_samples": 5000,
"train_samples_per_second": 37.964,
"train_steps_per_second": 0.297
}