mirror of
https://github.com/zenlm/zen-scribe.git
synced 2026-10-30 14:57:44 +00:00
zeekay
95ee5587a5
- training/identity/dataset.json: 49 identity Q&A examples for Zen Scribe (4B) - training/identity/dataset_info.json: alpaca format descriptor for zoo-gym - training/config.yaml: LoRA SFT config (rank 8, qwen3 template, bf16) - training/train.sh: zoo-gym training entrypoint script
42 lines
685 B
YAML
42 lines
685 B
YAML
### model
|
|
model_name_or_path: zenlm/zen-scribe
|
|
|
|
### method
|
|
stage: sft
|
|
do_train: true
|
|
finetuning_type: lora
|
|
lora_target: all
|
|
lora_rank: 8
|
|
lora_alpha: 16
|
|
lora_dropout: 0.05
|
|
|
|
### dataset
|
|
dataset: identity
|
|
dataset_dir: ./training/identity
|
|
template: qwen3
|
|
cutoff_len: 2048
|
|
max_samples: 100
|
|
overwrite_cache: true
|
|
|
|
### output
|
|
output_dir: ./training/output
|
|
save_steps: 50
|
|
logging_steps: 10
|
|
plot_loss: true
|
|
overwrite_output_dir: true
|
|
|
|
### train
|
|
per_device_train_batch_size: 4
|
|
gradient_accumulation_steps: 2
|
|
learning_rate: 1.0e-4
|
|
num_train_epochs: 3
|
|
lr_scheduler_type: cosine
|
|
warmup_ratio: 0.1
|
|
bf16: true
|
|
|
|
### eval
|
|
val_size: 0.1
|
|
per_device_eval_batch_size: 2
|
|
eval_strategy: steps
|
|
eval_steps: 50
|