| checkpointing_steps: null | |
| clip_grad_norm: -1 | |
| config_name: null | |
| dataset_config_name: null | |
| dataset_name: null | |
| gradient_accumulation_steps: 16 | |
| gradient_checkpointing: false | |
| learning_rate: 2.0e-05 | |
| logging_steps: 1 | |
| lora_alpha: 16 | |
| lora_dropout: 0.1 | |
| lora_rank: 64 | |
| low_cpu_mem_usage: false | |
| lr_scheduler_type: linear | |
| max_seq_length: 1024 | |
| max_train_steps: 868 | |
| model_name_or_path: /net/nfs/mosaic/day/llama_hf/llama-2-13b | |
| num_train_epochs: 2 | |
| output_dir: /net/nfs/mosaic/day/uniagent/train/output/lumos_unified_ground_iterative-13B_new/ | |
| overwrite_cache: false | |
| per_device_train_batch_size: 2 | |
| preprocessing_num_workers: 16 | |
| report_to: tensorboard | |
| resume_from_checkpoint: null | |
| seed: null | |
| tokenizer_name: /net/nfs/mosaic/day/llama_hf/llama-2-13b | |
| train_file: data/train/unified/train_annots/lumos_unified_ground_iterative.jsonl | |
| use_8bit_optimizer: false | |
| use_flash_attn: true | |
| use_lora: false | |
| use_qlora: false | |
| use_slow_tokenizer: true | |
| warmup_ratio: 0.03 | |
| weight_decay: 0.0 | |
| with_tracking: true | |