How to use deepspeed zero3 with cpu offload?

[l3utterfly]

Please check that this issue hasn't been reported before.

• [X] I searched previous Bug Reports didn't find any similar reports.

Expected Behavior

Working

Current behaviour

Using the following config:

{
  "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,
    "offload_optimizer": {
      "device": "cpu",
      "pin_memory": true
    },
    "offload_param": {
      "device": "cpu",
      "pin_memory": true
    },
    "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
}

I get eval loss = nan or loss = 0 after a few steps of training.

Steps to reproduce

1. Mistral 7B model full tune
2. Open hermes 2.5 dataset

Config yaml

base_model: /home/layla/src/text-generation-webui/models/Mistral-7B-v0.1
base_model_config: /home/layla/src/text-generation-webui/models/Mistral-7B-v0.1
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: /home/layla/src/Layla-datasets/datasets_formatted/base/dailydialog.topicalchat.openhermes.jsonl
    ds_type: json # see other options below
    type: sharegpt
    conversation: vicuna_v1.1

# datasets:
#   - path: /home/layla/src/Layla-datasets/datasets_formatted/airoboros_alpaca.jsonl
#     type: alpaca

dataset_prepared_path: last_run_prepared
val_set_size: 0.002
output_dir: ./out

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

wandb_project:
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0000005

train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_ratio: 0.05
eval_steps: 0.1
eval_table_size:
eval_table_max_new_tokens: 128
eval_sample_packing: false
save_steps: 100
debug:
deepspeed: /home/layla/src/Layla-datasets/axolotl/configs/deepspeed/zero3_cpuoffload.json # multi-gpu only
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

Possible solution

No response

Which Operating Systems are you using?

• [X] Linux
• [ ] macOS
• [ ] Windows

Python Version

3.11

axolotl branch-commit

2d65f470d54986f2df542cb8332e31113c04112d

Acknowledgements

• [X] My issue title is concise, descriptive, and in title casing.
• [X] I have searched the existing issues to make sure this bug has not been reported yet.
• [X] I am using the latest version of axolotl.
• [X] I have provided enough information for the maintainers to reproduce and diagnose the issue.

[NanoCode012]

Are you sure it's due to cpu offload? Can you try use the regular ds3_bf16