| data_root_dir: /hai/scratch/belkhale/datasets | |
| hf_token: .hf_token | |
| image_aug: false | |
| is_resume: true | |
| pretrained_checkpoint: runs/prism-qwen25-dinosiglip-224px+0_5b+mx-libero-90+n1+b32+x7--extra_tokens/checkpoints/step-060000-epoch-26-loss=0.1209.pt | |
| resume_epoch: 26 | |
| resume_step: 60000 | |
| run_id: prism-qwen25-dinosiglip-224px+0_5b+mx-libero-90+n1+b32+x7--extra_tokens | |
| run_id_note: extra_tokens | |
| run_root_dir: runs | |
| save_interval: 2500 | |
| seed: 7 | |
| trackers: | |
| - jsonl | |
| - wandb | |
| vla: | |
| action_tokenizer: extra_action_tokenizer | |
| base_vlm: prism-qwen25-extra-dinosiglip-224px+0_5b | |
| data_mix: libero_90 | |
| enable_gradient_checkpointing: true | |
| enable_mixed_precision_training: true | |
| epochs: 1000 | |
| expected_world_size: 8 | |
| freeze_llm_backbone: false | |
| freeze_vision_backbone: false | |
| global_batch_size: 256 | |
| learning_rate: 2.0e-05 | |
| lr_scheduler_type: constant | |
| max_grad_norm: 1.0 | |
| max_steps: null | |
| per_device_batch_size: 32 | |
| reduce_in_full_precision: true | |
| save_every_n_steps: 25000 | |
| shuffle_buffer_size: 256000 | |
| train_strategy: fsdp-full-shard | |
| type: prism-qwen25-dinosiglip-224px+0_5b+mx-libero-90 | |
| unfreeze_last_llm_layer: false | |
| vla_id: prism-qwen25-dinosiglip-224px+0_5b+mx-libero-90 | |
| warmup_ratio: 0.0 | |
| weight_decay: 0.0 | |
| wandb_entity: null | |
| wandb_project: prismatic | |