stage string | do_train bool | model_name_or_path string | dataset string | template string | finetuning_type string | lora_target string | output_dir string | per_device_train_batch_size int64 | gradient_accumulation_steps int64 | lr_scheduler_type string | logging_steps int64 | warmup_ratio float64 | save_steps int64 | learning_rate float64 | num_train_epochs float64 | max_samples int64 | max_grad_norm float64 | loraplus_lr_ratio float64 | fp16 bool | report_to string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
sft | true | Qwen/Qwen3-4B | formatted_dialogs | qwen3 | lora | all | ./qwen3_4B_lora | 4 | 4 | cosine | 5 | 0.1 | 5,000 | 0.00005 | 3 | 10,000 | 1 | 16 | true | none |
No dataset card yet
- Downloads last month
- 2