bf16: true
cutoff_len: 16384
dataset: mlfoundations-dev/r1_annotated_math
dataset_dir: ONLINE
ddp_timeout: 180000000
deepspeed: dcft/train/zero3.json
do_train: true
eval_strategy: 'no'
finetuning_type: full
formatting: sharegpt
global_batch_size: 96
gradient_accumulation_steps: 3
hub_model_id: mlfoundations-dev/llama3-1_8b_r1_annotated_math
include_hp: dcft/train/hp_settings/reasoning.yaml
learning_rate: 1.0e-05
logging_steps: 1
lr_scheduler_type: cosine
max_samples: 1000000
messages: conversations
model_name_or_path: Qwen/Qwen2.5-7B-Instruct
num_train_epochs: 3.0
output_dir: /tmp/dcft_checkpoints/llama3-1_8b_r1_annotated_math
overwrite_cache: true
per_device_train_batch_size: 1
plot_loss: true
preprocessing_num_workers: 16
push_to_db: true
push_to_hub: true
report_to: wandb
run_name: llama3-1_8b_r1_annotated_math
save_strategy: epoch
stage: sft
template: qwen25
warmup_ratio: 0.1