AI开发平台ModelArts-Yaml配置文件参数配置说明:rm_yaml样例模板

时间:2025-02-12 15:14:13

rm_yaml样例模板

### modelmodel_name_or_path: /home/ma-user/ws/tokenizers/llama3-8b### methodstage: rmdo_train: true# 全参# finetuning_type: full# lorafinetuning_type: loralora_target: alldeepspeed: examples/deepspeed/ds_z0_config.json### datasetdataset: dpo_en_demotemplate: llama3cutoff_len: 4096max_samples: 50000overwrite_cache: truepreprocessing_num_workers: 16dataloader_num_workers: 8packing: true### outputoutput_dir: /home/ma-user/ws/saves/rm/llama3-8b/loralogging_steps: 1save_steps: 500plot_loss: trueoverwrite_output_dir: true### trainper_device_train_batch_size: 1gradient_accumulation_steps: 8learning_rate: 1.0e-4num_train_epochs: 3.0lr_scheduler_type: cosinewarmup_ratio: 0.1bf16: trueddp_timeout: 180000000include_tokens_per_second: trueinclude_num_input_tokens_seen: true
support.huaweicloud.com/bestpractice-modelarts/modelarts_llm_train_91040.html