| !!python/object:icrt.util.args.ExperimentConfig |
| dataset_cfg: !!python/object:icrt.util.args.DatasetConfig |
| action_noise: 0.0 |
| action_traj_noise: false |
| dataset_fraction: 1.0 |
| dataset_json: |
| - /storage/nfs/xzhang/experiments/nopool_exps/exp_ds8_libero_1x_task_robocasa_mem_eight_GBS256_cfgtask_robocasa_mem_eight_sl2048_s2_crossmae_block_sigmoid_g1_nopool_compress/exp_ds8_libero_1x_task_robocasa_mem_eight_GBS256_cfgtask_robocasa_mem_eight_sl2048_s2_crossmae_block_sigmoid_g1_nopool_compress/task_robocasa_mem_eight.json |
| dataset_val_json: |
| - '' |
| goal_conditioned: false |
| load_in_mem: false |
| low_dim_noise: 0.005 |
| min_subsample_length: 64 |
| n_examples_only: -1 |
| non_overlapping: |
| - 2 |
| num_repeat_traj: |
| - 256 |
| num_weighted_steps: 30 |
| only_k: 10 |
| proprio_noise: 0.005 |
| random_patch_masking: false |
| rebalance_indices: false |
| rebalance_tasks: true |
| sample_each_state: false |
| shuffle_repeat_traj: true |
| shuffle_rnd_prob: 0.0 |
| sim_score_th: -1.0 |
| skip_step: false |
| sort_by_closeness: false |
| sort_by_lang: true |
| start_from_beginning: false |
| subsample_traj: false |
| task_barrier: true |
| task_name: '' |
| task_names: null |
| toy_max_distractors: 1 |
| toy_max_tasks: 5 |
| toy_task_id: 1 |
| toy_train_dataset_size: 1000000 |
| toy_train_seed: 42 |
| toy_val_dataset_size: 500000 |
| toy_val_seed: 84 |
| train_on_exploration: false |
| trim_num: -1 |
| trim_set: '' |
| use_dali: true |
| val_only_k: 5 |
| video_conditioned: false |
| vision_aug: true |
| weight_by_dataset: |
| - 1 |
| dataset_json_contents: |
| - action_keys: |
| - actions |
| dataset_path: |
| - memory/MemPutKBreadInMicrowave/2025-07-25-10-40-55/demo_im128_notp.hdf5 |
| - memory/MemPutKBreadInMicrowave/2025-07-25-13-25-05/demo_im128_notp.hdf5 |
| - memory/MemPutKBreadInMicrowave/2025-07-25-14-50-59/demo_im128_notp.hdf5 |
| - memory/MemFruitInSinkLeftFar/2025-07-13-15-44-32/demo_im128_notp.hdf5 |
| - memory/MemFruitInSinkRightFar/2025-07-15-15-07-57/demo_im128_notp.hdf5 |
| - memory/MemHeatPot/2025-07-24-22-26-20/demo_im128_notp.hdf5 |
| - memory/MemHeatPot/2025-07-25-15-56-53/demo_im128_notp.hdf5 |
| - memory/MemWashAndReturnLeft/2025-07-25-00-12-14/demo_im128_notp.hdf5 |
| - memory/MemWashAndReturnRight/2025-07-25-00-45-53/demo_im128_notp.hdf5 |
| - memory/MemHeatPotMultiple/2025-09-20-20-10-55/demo_im128_notp.hdf5 |
| - memory/MemHeatPotMultiple/2025-09-20-20-35-24/demo_im128_notp.hdf5 |
| - memory/MemHeatPotMultiple/2025-09-20-20-50-23/demo_im128_notp.hdf5 |
| - memory/MemHeatPotMultiple/2025-09-20-21-08-42/demo_im128_notp.hdf5 |
| - memory/MemHeatPotMultiple/2025-09-20-21-35-46/demo_im128_notp.hdf5 |
| - memory/MemPutKBowlInCabinet/seed0_2025-09-19-23-10-47_6demo/demo_im128_notp.hdf5 |
| - memory/MemPutKBowlInCabinet/seed1_2025-09-19-23-32-29_20demo/demo_im128_notp.hdf5 |
| - memory/MemPutKBowlInCabinet/seed2_2025-09-20-00-42-49_30demo/demo_im128_notp.hdf5 |
| - memory/MemRetrieveOilsFromCounterLL/2025-09-20-22-49-27/demo_im128_notp.hdf5 |
| - memory/MemRetrieveOilsFromCounterLR/2025-09-20-23-25-19/demo_im128_notp.hdf5 |
| - memory/MemRetrieveOilsFromCounterRL/2025-09-20-23-06-15/demo_im128_notp.hdf5 |
| - memory/MemRetrieveOilsFromCounterRR/2025-09-20-23-14-26/demo_im128_notp.hdf5 |
| - memory/MemWashAndReturnSameLocation/2025-09-20-21-52-31/demo_im128_notp.hdf5 |
| - memory/MemWashAndReturnSameLocation/2025-09-20-22-11-00/demo_im128_notp.hdf5 |
| - memory/MemWashAndReturnSameLocation/2025-09-20-22-25-19/demo_im128_notp.hdf5 |
| group_method: dinov2 |
| image_keys: |
| - obs/robot0_agentview_center_image |
| - obs/robot0_eye_in_hand_image |
| low_dim_keys: [] |
| max_traj_len: 2048 |
| proprio_keys: |
| - obs/robot0_joint_pos_cos |
| - obs/robot0_gripper_qpos |
| device: cuda |
| dist_on_itp: false |
| dist_url: env:// |
| load_config: null |
| local_rank: -1 |
| logging_cfg: !!python/object:icrt.util.args.LoggingConfig |
| log_dir: /storage/nfs/xzhang/experiments/nopool_exps/exp_ds8_libero_1x_task_robocasa_mem_eight_GBS256_cfgtask_robocasa_mem_eight_sl2048_s2_crossmae_block_sigmoid_g1_nopool_compress/exp_ds8_libero_1x_task_robocasa_mem_eight_GBS256_cfgtask_robocasa_mem_eight_sl2048_s2_crossmae_block_sigmoid_g1_nopool_compress |
| log_name: exp_ds8_libero_1x_task_robocasa_mem_eight_GBS256_cfgtask_robocasa_mem_eight_sl2048_s2_crossmae_block_sigmoid_g1_nopool_compress |
| output_dir: /storage/nfs/xzhang/experiments/nopool_exps/exp_ds8_libero_1x_task_robocasa_mem_eight_GBS256_cfgtask_robocasa_mem_eight_sl2048_s2_crossmae_block_sigmoid_g1_nopool_compress/exp_ds8_libero_1x_task_robocasa_mem_eight_GBS256_cfgtask_robocasa_mem_eight_sl2048_s2_crossmae_block_sigmoid_g1_nopool_compress |
| model_cfg: !!python/object:icrt.util.args.ModelConfig |
| policy_cfg: !!python/object:icrt.util.args.PolicyConfig |
| adapter_mlp_ratio: 4.0 |
| adapter_num_heads: 8 |
| add_sink_token_to_full_attn: false |
| attn_drop: 0.0 |
| attn_latent_len: 1 |
| attn_scale_factor: null |
| block_attn_ind: null |
| block_pattern_start_offset: 0 |
| compress_full_attn: true |
| compress_include_prev_action: false |
| compressor_latent_len: 1 |
| decoder_cfg_path: config/model_config/action_head.json |
| decoder_pred_head: mlp |
| full_attn_inds: null |
| gate_full_attn_layers: true |
| gating_type: sigmoid_g1 |
| gr00t_add_dit_lora: false |
| gr00t_add_full_attn: true |
| gr00t_add_lora_to_vision_projector: false |
| gr00t_compute_dtype: bfloat16 |
| gr00t_full_attn_idx: |
| - 3 |
| - 7 |
| - 11 |
| - 15 |
| gr00t_pretrained_model: true |
| gr00t_tune_diffusion_model: true |
| gr00t_tune_llm: false |
| gr00t_tune_projector: true |
| gr00t_tune_vision_projector: false |
| gr00t_tune_visual: false |
| hist_attn_latent_len: 1 |
| hist_blind_gating: false |
| hist_feature_gating: false |
| interpolation_factor: null |
| llama_ckpt_dir: /home/rutavms/.llama/checkpoints/Llama3.2-1B-Instruct |
| load_llama: true |
| lora_layer_idxs: null |
| lora_rank: 4 |
| loss_w_action: 1.0 |
| model_version: v2 |
| multikv_attn_pool: false |
| no_prompt_loss: true |
| num_inference_diffusion_steps: null |
| num_train_diffusion_steps: null |
| phase: pretrain |
| pool_block_inds: null |
| position_language_embedding: start |
| pretrained_path: null |
| prop_act_norm: null |
| remove_action: false |
| remove_proprio: false |
| scale_loss: 1.0 |
| scratch_llama_config: config/model_config/libero_1x.json |
| separate_camera_adapter: true |
| skip_obs_pooling: true |
| step_weight: 1.0 |
| tt_window_obs: false |
| use_block_attention: true |
| use_linear_attention: false |
| use_lstm: false |
| use_mamba: false |
| use_reshape_block_attn: false |
| use_transformer_xl: false |
| use_yarn: null |
| vision_projection_layer: false |
| xl_chunk_factor: 32 |
| xl_mem_len: 32 |
| vision_encoder_cfg: !!python/object:icrt.util.args.VisionEncoderConfig |
| bbox_vision_encoder_cfg: config/model_config/bbox_vit_128_16.json |
| freeze_all: false |
| vision_encoder: /storage/nfs/rshah/datasets/robocasa/datasets/crossmae_rtx/cross-mae-rtx-vitb.pth |
| vision_lora: false |
| vision_lora_rank: 8 |
| vision_nonpretrained: false |
| vision_unfreeze_all: false |
| vision_unfreeze_last_n: 0 |
| optimizer_cfg: !!python/object:icrt.util.args.OptimizerConfig |
| blr: 0.001 |
| lr: 0.0005 |
| min_lr: 0.0 |
| warmup_epochs: 2.0 |
| weight_decay: 0.01 |
| shared_cfg: !!python/object:icrt.util.args.SharedConfig |
| batch_size: 2 |
| bbox_noise: 0.015625 |
| block_finetune: '' |
| downsample_obs: 8 |
| drop_bbox_prob: 0.0 |
| dropping_prompt_rnd: false |
| enable_gradient_checkpointing: false |
| gr00t_embodiment_str: single_panda_omron |
| gr00t_metadata_path: metadata.json |
| gr00t_model_path: nvidia/GR00T-N1.5-3B |
| gripper_img: true |
| has_base_action: true |
| hist_seq_length: 2048 |
| is_bimanual: false |
| k_ptp: 0 |
| max_bbox_per_img: 15 |
| no_img: false |
| no_img_low_dim: 42 |
| normalize_io: false |
| num_cameras: 2 |
| num_pred_steps: 32 |
| num_stages: 1 |
| only_first_obs: false |
| pad_to_max_length: true |
| resume: null |
| rot_6d: false |
| rot_euler: false |
| save_every: 10 |
| seed: 2 |
| seq_length: 2048 |
| split_epoch: 1 |
| start_epoch: 0 |
| state_center_coords: false |
| tokenizer_path: physical-intelligence/fast |
| use_bboxes: false |
| use_delta_action: false |
| use_fast_tokenizer: false |
| use_gr00t: false |
| use_history: false |
| use_language_conditioning: true |
| use_toy_vision_dataset: false |
| use_vqvae_tokenizer: false |
| train: true |
| trainer_cfg: !!python/object:icrt.util.args.TrainerConfig |
| accum_iter: 32 |
| break_after_n_epochs: 101 |
| compile_model: true |
| epochs: 200 |
| num_workers: 72 |
| pin_memory: true |
| val_every: 10 |
| wandb_project: icrt_reruns |
| wandb_watch: true |
| world_size: 1 |
|
|
