chunked/nltk_chunking/_deepspeed/chunk_117.txt

yaml
{
   "scheduler": {
         "type": "WarmupDecayLR",
         "params": {
             "total_num_steps": "auto",
             "warmup_min_lr": "auto",
             "warmup_max_lr": "auto",
             "warmup_num_steps": "auto"
         }
     }
}

Precision
Deepspeed supports fp32, fp16, and bf16 mixed precision.