adam_beta1: 0.9
adam_beta2: 0.999
bf16: true
cutoff_len: 2048
dataset: mlfoundations-dev/oh-dcft-v3.1-llama-3.1-nemotron-70b
dataset_dir: ONLINE
ddp_timeout: 180000000
deepspeed: dcft/train/zero3.json
do_train: true
enable_liger_kernel: false
eval_strategy: epoch
finetuning_type: full
formatting: sharegpt
global_batch_size: 512
gradient_accumulation_steps: 1
gradient_checkpointing: true
hub_model_id: mlfoundations-dev/mistral_7b_0-3_oh-dcft-v3.1-llama-3.1-nemotron-70b
include_hp: dcft/train/hp_settings/mistral_sft.yaml
learning_rate: 5.0e-06
logging_steps: 10
lr_scheduler_kwargs:
  min_lr: 5.0e-07
lr_scheduler_type: cosine_with_min_lr
max_grad_norm: 1
messages: conversations
model_name_or_path: mistralai/Mistral-7B-v0.3
neat_packing: true
num_train_epochs: 3.0
output_dir: /tmp/experiments/train/checkpoints/oh-dcft-v3.1-llama-3.1-nemotron-70b
overwrite_cache: true
overwrite_output_dir: true
packing: true
per_device_train_batch_size: 16
plot_loss: true
preprocessing_num_workers: 16
push_to_db: true
push_to_hub: true
report_to: wandb
run_name: mistral_7b_0-3_oh-dcft-v3.1-llama-3.1-nemotron-70b
save_strategy: epoch
stage: sft
template: mistral
val_size: 0.05
warmup_ratio: 0.05
weight_decay: 0.1