| config: | |
| lm_model_name: meta-llama/Meta-Llama-3-8B-Instruct | |
| lr: 0.0001 | |
| weight_decay: 0.01 | |
| warmup_steps: 100 | |
| vision_hidden_size: 1024 | |
| freeze_vision: true | |
| use_lora: true | |
| vision_layer: -1 | |
| num_workers: 8 | |
| val_num_workers: 1 | |
| shuffle_buffer: 10000 | |
| resampled: false | |
| max_length: 32 | |
| training_epochs: 1 | |
| batch_size: 16 | |
| devices: 2 | |
| lora_r: 8 | |
| lora_alpha: 16 | |
| lora_dropout: 0.05 | |
| num_sanity_val_steps: 2 | |
| val_check_interval: 1000 | |
| limit_val_batches: 16 | |
| num_training_step: 68750 | |