llm / examples /train_full /dolci_10k_with_tool_call_batch3.yaml
dongxx1104's picture
Upload folder using huggingface_hub
db704cb verified
### model
model_name_or_path: Qwen/Qwen3-8B-Base
trust_remote_code: true
### method
stage: sft
do_train: true
finetuning_type: full
deepspeed: /shared_workspace_mfs/ximing/LLaMA-Factory/examples/deepspeed/ds_z3_offload_config.json
# choices: [ds_z0_config.json, ds_z2_config.json, ds_z3_config.json, ds_z3_offload_config.json]
### dataset
dataset: dolci_10k_with_tool_call_batch3
template: qwen3_nothink
cutoff_len: 65536
rope_scaling: yarn
max_samples: 10000
overwrite_cache: true
preprocessing_num_workers: 16
dataloader_num_workers: 4
### output
output_dir: /shared_workspace_mfs/ximing/sft_dolci_10k_with_tool_call_batch3
logging_steps: 1
save_strategy: epoch
plot_loss: true
overwrite_output_dir: true
report_to: wandb
### train
per_device_train_batch_size: 4
gradient_accumulation_steps: 4
learning_rate: 5e-5
num_train_epochs: 1
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000
resume_from_checkpoint: null
### eval
# eval_dataset: alpaca_en_demo
# val_size: 0.1
# per_device_eval_batch_size: 1
# eval_strategy: steps
# eval_steps: 500