# Resolved R13 LIBERO Goal+Long posttraining configuration. # Package/import names are normalized to the public w0 implementation. # Private absolute paths are represented by portable environment variables; all other # training values are unchanged from the source experiment.yaml. output_dir: ${oc.env:W0_OUTPUT_DIR,outputs/libero/goal_long_r13/step_000600} batch_size: 48 num_workers: 4 gc_collect_every: 1 lr_scheduler_type: cosine learning_rate: 5.0e-05 soft_prompt_lr_scale: 1.0 num_epochs: 10 max_steps: null log_every: 10 save_every: 100 keep_training_states: 1 checkpoint_min_free_gib: 128 eval_every: 50 eval_num_inference_steps: 10 gradient_accumulation_steps: 1 mixed_precision: bf16 seed: 42 max_grad_norm: 1.0 weight_decay: 0.0 freeze_video_dit: false resume: null init_checkpoint: ${oc.env:W0_LIBERO_INIT_CHECKPOINT,checkpoints/libero_init_step_2640.pt} init_soft_prompts: false resume_global_batch_size: null wandb: enabled: false workspace: null project: w0 name: w0-lib-gl-rf-c16-e10-32g-r13 group: null mode: online data: train: _target_: w0.datasets.libero_posttrain.LiberoPosttrainDataset dataset_root: ${oc.env:W0_LIBERO_DATA,data/libero} is_training_set: true samples_per_epoch: 100000 epoch_sampling: random_frame seed: 42 val_set_proportion: 0.01 pretrained_norm_stats: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json} max_action_offset: 15 action_horizon: 64 action_chunk_size: 16 video_plan_horizon: 96 action_video_freq_ratio: 8 text_embedding_cache_dir: ${oc.env:W0_LIBERO_TEXT_CACHE,assets/libero/text_embeds_cache} include_suites: - libero_goal - libero_10 color_jitter: enabled: false brightness: 0.2 contrast: 0.2 saturation: 0.2 hue: 0.0 processor: _target_: w0.datasets.libero_posttrain.LiberoPosttrainProcessor stats_path: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json} val: _target_: w0.datasets.libero_posttrain.LiberoPosttrainDataset dataset_root: ${oc.env:W0_LIBERO_DATA,data/libero} is_training_set: false samples_per_epoch: 128 epoch_sampling: random_frame seed: 42 val_set_proportion: 0.01 pretrained_norm_stats: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json} max_action_offset: 15 action_horizon: 64 action_chunk_size: 16 video_plan_horizon: 96 action_video_freq_ratio: 8 text_embedding_cache_dir: ${oc.env:W0_LIBERO_TEXT_CACHE,assets/libero/text_embeds_cache} include_suites: - libero_goal - libero_10 processor: _target_: w0.datasets.libero_posttrain.LiberoPosttrainProcessor stats_path: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json} color_jitter: enabled: false brightness: 0.2 contrast: 0.2 saturation: 0.2 hue: 0.0 model: _target_: w0.runtime.create_w0 model_id: ${oc.env:W0_WAN_MODEL_DIR,assets/wan22} tokenizer_model_id: Wan-AI/Wan2.1-T2V-1.3B tokenizer_max_len: 128 load_text_encoder: false proprio_dim: 37 redirect_common_files: true mot_checkpoint_mixed_attn: true action_dit_pretrained_path: null skip_dit_load_from_pretrain: true legacy_checkpoint_config: null observation_encoder: dino dino_config: model_name: dinov3-vitb16-hf weights_path: ${oc.env:W0_DINO_WEIGHTS,assets/dinov3-vitb16-pretrain-lvd1689m} input_resolution: - 384 - 320 feature_dim: 768 patch_size: 16 use_cls_token: false normalize_features: false encode_microbatch_size: 16 checkpoint_observation_encoder_adapt: false action_horizon: 64 action_chunk_size: 16 max_action_offset: 15 action_train_timestep_mode: per_chunk future_training_mode: idm idm_video_cond_noise_prob: 0.5 idm_checkpoint_full_layer: true idm_checkpoint_layer_stride: 1 action_obs_downsample_factor: 1 chunk_kv_editor_num_queries: 32 num_history_frames: 0 action_video_read_mode: future video_rope_frame_stride: 1 video_dit_config: compact_timestep_embedding: true has_image_input: false patch_size: - 1 - 2 - 2 in_dim: 48 hidden_dim: 3072 ffn_dim: 14336 freq_dim: 256 text_dim: 4096 out_dim: 48 num_heads: 24 attn_head_dim: 128 num_layers: 30 eps: 1.0e-06 seperated_timestep: true require_clip_embedding: false require_vae_embedding: false fuse_vae_embedding_in_latents: true use_gradient_checkpointing: true video_attention_mask_mode: first_frame_causal action_conditioned: false action_dim: 37 action_group_causal_mask_mode: group_diagonal action_dit_config: num_soft_prompts: 8 soft_prompt_domains: - libero/franka soft_prompt_init_std: 0.02 domain_configs: - name: libero/franka dataset: libero embodiment: franka action_dim: 37 proprio_dim: 37 action_valid_indices: - 7 - 8 - 9 - 10 - 11 - 12 - 13 - 14 - 15 - 16 proprio_valid_indices: - 7 - 8 - 9 - 10 - 11 - 12 - 13 - 14 - 15 - 16 raw_eef_names: - eef_x - eef_y - eef_z - rot6d_col0_x - rot6d_col0_y - rot6d_col0_z - rot6d_col1_x - rot6d_col1_y - rot6d_col1_z - gripper_open_scale action_semantics: native_delta_eef10_row_aligned action_dim: 37 hidden_dim: 1024 ffn_dim: 4096 num_heads: 24 attn_head_dim: 128 num_layers: 30 text_dim: 4096 freq_dim: 256 eps: 1.0e-06 autoregressive_teacher_forcing: true action_chunk_size: 16 action_teacher_forcing_mask_mode: stage1_chunkwise use_gradient_checkpointing: true video_scheduler: train_shift: 5.0 infer_shift: 5.0 num_train_timesteps: 1000 action_scheduler: train_shift: 5.0 infer_shift: 5.0 num_train_timesteps: 1000 loss: lambda_action: 1.0 lambda_action_prior: 1.0 video_plan_horizon: 96 video_action_stride: 8 video_plan_steps: 10 domain_registry_path: null posttrain_required: true posttrain_stats: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json} history_aware_batching: false optimizer_group_max_numel: 200000000