adam_beta1: 0.9
adam_beta2: 0.999
assistant_tag: gpt
bf16: true
content_tag: value
cutoff_len: 2048
dataset: /p/data1/mmlaion/marianna/lf_datasets/mlfoundations-dev/oh-dcft-v3.1-gpt-4o-mini
dataset_dir: ONLINE
deepspeed: dcft/train/zero3.json
do_train: true
enable_liger_kernel: false
eval_strategy: 'no'
finetuning_type: full
formatting: sharegpt
global_batch_size: 2048
gradient_accumulation_steps: 2
gradient_checkpointing: true
hub_model_id: mlfoundations-dev/oh-mistral-bs2048_lr5_00E-06_schedulerconstant_warmup5_00E-02_minlr
learning_rate: 5.0e-06
logging_steps: 10
lr_scheduler_kwargs:
  min_lr: null
lr_scheduler_type: constant
max_grad_norm: 1.0
messages: conversations
model_name_or_path: mistralai/Mistral-7B-v0.3
neat_packing: true
num_train_epochs: 3.0
output_dir: /p/data1/mmlaion/marianna/dcft_checkpoints/oh-mistral-bs2048_lr5.00E-06_schedulerconstant_warmup5.00E-02_minlr
overwrite_cache: true
overwrite_output_dir: false
packing: true
per_device_train_batch_size: 8
plot_loss: false
preprocessing_num_workers: 32
push_to_db: false
push_to_hub: false
report_to: wandb
role_tag: from
run_name: oh-mistral-bs2048_lr5.00E-06_schedulerconstant_warmup5.00E-02_minlr
save_strategy: epoch
stage: sft
template: mistral
user_tag: human
warmup_ratio: 0.05
weight_decay: 0.1