deepspeed_config.json

{
    "train_batch_size": 8, 
    "gradient_accumulation_steps": 8,  
    "zero_optimization": {
      "stage": 2
    },
    "mixed_precision": "bf16",  
    "bf16": {
      "enabled": true
    },
    "fp16": {
      "enabled": false
    },
    "zero_allow_untested_optimizer": true
  }