[BUG]when using deepspeed loss learning rate did not down

Describe the bug
i want to train a dolly 2.0 2.8b model with using deepspeed but display on terminal is always same did i miss something?
without using deepspeed it can normal down

this is using deepspeed when training

this is without using deepspeed

and this is my deepspeed file

{
"fp16": {
"enabled": false
},
"bf16": {
"enabled": true
},
"optimizer": {
"type": "AdamW",
"params": {
"lr": "auto",
"betas": "auto",
"eps": "auto",
"weight_decay": "auto"
}
},
"scheduler": {
"type": "WarmupLR",
"params": {
"warmup_min_lr": "auto",
"warmup_max_lr": "auto",
"warmup_num_steps": "auto"
}
},
"zero_optimization": {
"stage": 3,
"overlap_comm": true,
"contiguous_gradients": true,
"sub_group_size": 1e9,
"reduce_bucket_size": "auto",
"stage3_prefetch_bucket_size": "auto",
"stage3_param_persistence_threshold": "auto",
"stage3_max_live_parameters": 1e9,
"stage3_max_reuse_distance": 1e9,
"stage3_gather_16bit_weights_on_model_save": true
},
"gradient_accumulation_steps": "auto",
"gradient_clipping": "auto",
"steps_per_print": 2000,
"train_batch_size": "auto",
"train_micro_batch_size_per_gpu": "auto",
"wall_clock_breakdown": false
}

its make me so confuse
please help me QQ

Start by reproducing Dolly 2.0 2.8b training with and without the supplied DeepSpeed configuration, then compare the loss and learning-rate behavior. Check the optimizer, WarmupLR scheduler, BF16 setting, and ZeRO stage 3 configuration; done means the cause is identified and the behavior is fixed or clearly documented.