diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s0__2026-05-20-05-16-00-cube-single-play-v0-fb-flowbc-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s0__2026-05-20-05-16-00-cube-single-play-v0-fb-flowbc-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..65457e3d11e5ea4d9eddadb8d9ba43b69e31aa74 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s0__2026-05-20-05-16-00-cube-single-play-v0-fb-flowbc-s0/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: pixels +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: true +use_wandb: true +wandb_project: factored-fb +wandb_run_name: visual-cube-single__fb_flowbc__s0__full1M_amsks +wandb_entity: amsks +wandb_group: null +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1.0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +frame_stack: 3 +obs_normalizer: + name: RGBNormalizerConfig +rgb_encoder: + name: drq + feature_dim: 256 +augmentator: + name: random_shifts + pad: 2 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s1__2026-05-20-05-40-13-cube-single-play-v0-fb-flowbc-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s1__2026-05-20-05-40-13-cube-single-play-v0-fb-flowbc-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7293754c4decc4824c5fb00eade945a9e9a57a17 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s1__2026-05-20-05-40-13-cube-single-play-v0-fb-flowbc-s1/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: pixels +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: true +use_wandb: true +wandb_project: factored-fb +wandb_run_name: visual-cube-single__fb_flowbc__s1__full1M_amsks +wandb_entity: amsks +wandb_group: null +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1.0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +frame_stack: 3 +obs_normalizer: + name: RGBNormalizerConfig +rgb_encoder: + name: drq + feature_dim: 256 +augmentator: + name: random_shifts + pad: 2 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s2__2026-05-20-05-40-16-cube-single-play-v0-fb-flowbc-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s2__2026-05-20-05-40-16-cube-single-play-v0-fb-flowbc-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cbc53db8b9d5f7e62f6d0cd9b4cb520604ab2313 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s2__2026-05-20-05-40-16-cube-single-play-v0-fb-flowbc-s2/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: pixels +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: true +use_wandb: true +wandb_project: factored-fb +wandb_run_name: visual-cube-single__fb_flowbc__s2__full1M_amsks +wandb_entity: amsks +wandb_group: null +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1.0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +frame_stack: 3 +obs_normalizer: + name: RGBNormalizerConfig +rgb_encoder: + name: drq + feature_dim: 256 +augmentator: + name: random_shifts + pad: 2 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s3__2026-05-20-05-40-19-cube-single-play-v0-fb-flowbc-s3/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s3__2026-05-20-05-40-19-cube-single-play-v0-fb-flowbc-s3/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..506188ad7e3acc32876769e3e9984bba7ef64141 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s3__2026-05-20-05-40-19-cube-single-play-v0-fb-flowbc-s3/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: pixels +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 3 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: true +use_wandb: true +wandb_project: factored-fb +wandb_run_name: visual-cube-single__fb_flowbc__s3__full1M_amsks +wandb_entity: amsks +wandb_group: null +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1.0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +frame_stack: 3 +obs_normalizer: + name: RGBNormalizerConfig +rgb_encoder: + name: drq + feature_dim: 256 +augmentator: + name: random_shifts + pad: 2 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s4__2026-05-20-05-40-22-cube-single-play-v0-fb-flowbc-s4/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s4__2026-05-20-05-40-22-cube-single-play-v0-fb-flowbc-s4/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ec795322b2fce7154ac1696ade176e65ac5a1628 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-pixels__s4__2026-05-20-05-40-22-cube-single-play-v0-fb-flowbc-s4/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: pixels +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 4 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: true +use_wandb: true +wandb_project: factored-fb +wandb_run_name: visual-cube-single__fb_flowbc__s4__full1M_amsks +wandb_entity: amsks +wandb_group: null +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1.0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +frame_stack: 3 +obs_normalizer: + name: RGBNormalizerConfig +rgb_encoder: + name: drq + feature_dim: 256 +augmentator: + name: random_shifts + pad: 2 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__20260607-062100-fb-state-encmatch-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__20260607-062100-fb-state-encmatch-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..79c6ea51f0d22799c3dfc1de649b22c58f29a99e --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__20260607-062100-fb-state-encmatch-s0/config.yaml @@ -0,0 +1,80 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_encmatch_s0 +wandb_entity: amsks +wandb_group: fb_state_encmatch +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +rgb_encoder: + name: mlp + feature_dim: 256 + hidden_dim: 1100 + hidden_layers: 5 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__20260607-070225-fb-state-plain-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__20260607-070225-fb-state-plain-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..702bc6133fcf3438592062aff9012d9cde8ff811 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__20260607-070225-fb-state-plain-s0/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s0 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__20260618-083115-fb-iql-ontraj-flow-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__20260618-083115-fb-iql-ontraj-flow-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b5fa480267edddabbaa19b29f0cd140dd1f0c398 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__20260618-083115-fb-iql-ontraj-flow-s0/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_iql_ontraj_flow_s0 +wandb_entity: amsks +wandb_group: fb_iql_ontraj_flow +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 1 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__a0-cubexyz-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__a0-cubexyz-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1f5d63e7ed79b6835bf045973a64799ae7240cce --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__a0-cubexyz-s0/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_a0_cubexyz_s0 +wandb_entity: amsks +wandb_group: fb_bridge_a0_cubexyz +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: false +d_norm: false +critic_goal_mix: +- 1.0 +- 0.0 +- 0.0 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: cube_xyz +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__decpl-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__decpl-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..56e07dbe8c07354425fc379cd84ab5bb507c7cf3 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__decpl-s0/config.yaml @@ -0,0 +1,89 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_decouple_s0 +wandb_entity: amsks +wandb_group: fb_cube_decouple +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: perm +decouple_actor_z: true +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__diql-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__diql-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..017db15dfcf917a850187c1db7f93c69c92ef079 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__diql-s0/config.yaml @@ -0,0 +1,89 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_decouple_iql_s0 +wandb_entity: amsks +wandb_group: fb_cube_decouple_iql +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: true +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__dzher-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__dzher-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c692f40390fff64d590709d9268239e342df3fc8 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__dzher-s0/config.yaml @@ -0,0 +1,89 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_decouple_zher_traj_s0 +wandb_entity: amsks +wandb_group: fb_cube_decouple_zher_traj +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: true +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__herlb-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__herlb-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..23a4827d0e3ac14ebfcb5e28497f9f678040adb8 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__herlb-s0/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_her_learnedB_r1_s0 +wandb_entity: amsks +wandb_group: fb_cube_her_learnedB_r1 +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__hernrm-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__hernrm-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f6132e8ce65657210699f37c7bdea07cde655bfe --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__hernrm-s0/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_her_idnorm_r1_s0 +wandb_entity: amsks +wandb_group: fb_cube_her_idnorm_r1 +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: identity_norm +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__iqllb-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__iqllb-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..61196a2a86c57ff99a03255c55c970402f6c71a3 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__iqllb-s0/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_iql_learnedB_r1_s0 +wandb_entity: amsks +wandb_group: fb_cube_iql_learnedB_r1 +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r0-base-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r0-base-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..257c416550a8e5096b2f21ef50b7765ed632307c --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r0-base-s0/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r0_base_s0 +wandb_entity: amsks +wandb_group: fb_bridge_r0_base +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: false +d_norm: false +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r1-decond-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r1-decond-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3d7b5606574a0dec8f891f3d663e35ca4ecb570c --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r1-decond-s0/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r1_decond_s0 +wandb_entity: amsks +wandb_group: fb_bridge_r1_decond +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: true +d_norm: false +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r2-dnorm-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r2-dnorm-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b7445dfafac83d49b0c441746a8dfe61916477a7 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r2-dnorm-s0/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r2_dnorm_s0 +wandb_entity: amsks +wandb_group: fb_bridge_r2_dnorm +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: false +d_norm: true +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r3-allon-s0/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r3-allon-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..666c7e3e5a08922eba02cea3cf37af5780cb6f16 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s0__r3-allon-s0/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r3_allon_s0 +wandb_entity: amsks +wandb_group: fb_bridge_r3_allon +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: true +d_norm: true +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__20260607-062100-fb-state-encmatch-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__20260607-062100-fb-state-encmatch-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d8dbd533b01af8ca5b39f781bd2a3df4548e73ca --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__20260607-062100-fb-state-encmatch-s1/config.yaml @@ -0,0 +1,80 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_encmatch_s1 +wandb_entity: amsks +wandb_group: fb_state_encmatch +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +rgb_encoder: + name: mlp + feature_dim: 256 + hidden_dim: 1100 + hidden_layers: 5 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__20260607-070225-fb-state-plain-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__20260607-070225-fb-state-plain-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..aef245e72fbe9afad1e8ca0d85a871adb462ad72 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__20260607-070225-fb-state-plain-s1/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s1 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__20260618-083115-fb-iql-ontraj-flow-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__20260618-083115-fb-iql-ontraj-flow-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b83ae38056a606607a6cbd6dea2b707882dfd0be --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__20260618-083115-fb-iql-ontraj-flow-s1/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_iql_ontraj_flow_s1 +wandb_entity: amsks +wandb_group: fb_iql_ontraj_flow +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 1 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__a0-cubexyz-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__a0-cubexyz-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4db473d6ec6be0dfc31fb93e7384fc18b6a68fb4 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__a0-cubexyz-s1/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_a0_cubexyz_s1 +wandb_entity: amsks +wandb_group: fb_bridge_a0_cubexyz +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: false +d_norm: false +critic_goal_mix: +- 1.0 +- 0.0 +- 0.0 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: cube_xyz +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__decpl-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__decpl-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e4dce9c5a0c7911db9cf8bdb6569d5afcbec14a4 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__decpl-s1/config.yaml @@ -0,0 +1,89 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_decouple_s1 +wandb_entity: amsks +wandb_group: fb_cube_decouple +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: perm +decouple_actor_z: true +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__diql-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__diql-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b7786b29a08571382160c4da54688062e4f60f98 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__diql-s1/config.yaml @@ -0,0 +1,89 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_decouple_iql_s1 +wandb_entity: amsks +wandb_group: fb_cube_decouple_iql +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: true +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__dzher-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__dzher-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c05e147e6fc5bc85812c80b5ac9066cbf3c87d9c --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__dzher-s1/config.yaml @@ -0,0 +1,89 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_decouple_zher_traj_s1 +wandb_entity: amsks +wandb_group: fb_cube_decouple_zher_traj +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: true +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__herlb-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__herlb-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4155027e3a67f61b3a67261c7f4dff96fa05f5a8 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__herlb-s1/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_her_learnedB_r1_s1 +wandb_entity: amsks +wandb_group: fb_cube_her_learnedB_r1 +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__hernrm-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__hernrm-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..640e7dfbb3bd1eb54906a9b9f0329a7ef1b14cb5 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__hernrm-s1/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_her_idnorm_r1_s1 +wandb_entity: amsks +wandb_group: fb_cube_her_idnorm_r1 +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: identity_norm +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__iqllb-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__iqllb-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3a9d96ee76322a6b5785beee1805d5050301da64 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__iqllb-s1/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_iql_learnedB_r1_s1 +wandb_entity: amsks +wandb_group: fb_cube_iql_learnedB_r1 +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r0-base-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r0-base-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12b6ec55aa2895f04ee68d6ba9aa6a5625b6d0af --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r0-base-s1/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r0_base_s1 +wandb_entity: amsks +wandb_group: fb_bridge_r0_base +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: false +d_norm: false +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r1-decond-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r1-decond-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ace075030a4a6c7d8d4c90a079cfbe98150c4618 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r1-decond-s1/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r1_decond_s1 +wandb_entity: amsks +wandb_group: fb_bridge_r1_decond +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: true +d_norm: false +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r2-dnorm-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r2-dnorm-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c4bcf94d3fa0ce11b1fd32a04db8415c2bd7deec --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r2-dnorm-s1/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r2_dnorm_s1 +wandb_entity: amsks +wandb_group: fb_bridge_r2_dnorm +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: false +d_norm: true +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r3-allon-s1/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r3-allon-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4b19fab5aa9b7869ef9b21ddbdd33686a57b19bb --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s1__r3-allon-s1/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r3_allon_s1 +wandb_entity: amsks +wandb_group: fb_bridge_r3_allon +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: true +d_norm: true +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__20260607-062100-fb-state-encmatch-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__20260607-062100-fb-state-encmatch-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fc452fbe85182fcab35849582b500aec3dff10b5 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__20260607-062100-fb-state-encmatch-s2/config.yaml @@ -0,0 +1,80 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_encmatch_s2 +wandb_entity: amsks +wandb_group: fb_state_encmatch +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +rgb_encoder: + name: mlp + feature_dim: 256 + hidden_dim: 1100 + hidden_layers: 5 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__20260607-070225-fb-state-plain-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__20260607-070225-fb-state-plain-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0c7256d49c79d93c3ddf0efa374c3c05efb0492c --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__20260607-070225-fb-state-plain-s2/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s2 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__20260618-083115-fb-iql-ontraj-flow-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__20260618-083115-fb-iql-ontraj-flow-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..77f104f3e76c312e653412854f62310cb1df8b21 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__20260618-083115-fb-iql-ontraj-flow-s2/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_iql_ontraj_flow_s2 +wandb_entity: amsks +wandb_group: fb_iql_ontraj_flow +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 1 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__a0-cubexyz-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__a0-cubexyz-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..652b0495879a79bb2474c0f611298e022df9c4e7 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__a0-cubexyz-s2/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_a0_cubexyz_s2 +wandb_entity: amsks +wandb_group: fb_bridge_a0_cubexyz +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: false +d_norm: false +critic_goal_mix: +- 1.0 +- 0.0 +- 0.0 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: cube_xyz +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__decpl-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__decpl-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..91d9281f00bbcf48eae13b8623e24d251e0db8f4 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__decpl-s2/config.yaml @@ -0,0 +1,89 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_decouple_s2 +wandb_entity: amsks +wandb_group: fb_cube_decouple +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: perm +decouple_actor_z: true +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__diql-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__diql-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..af468998310caf46c5e29d28c168632b1b4e5584 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__diql-s2/config.yaml @@ -0,0 +1,89 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_decouple_iql_s2 +wandb_entity: amsks +wandb_group: fb_cube_decouple_iql +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: true +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__dzher-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__dzher-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..32f0e7fedda1b6c3784729bcd047ddf427f9686c --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__dzher-s2/config.yaml @@ -0,0 +1,89 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_decouple_zher_traj_s2 +wandb_entity: amsks +wandb_group: fb_cube_decouple_zher_traj +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: true +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__herlb-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__herlb-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eb9e3a549493f86237d78558776b237c160559e7 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__herlb-s2/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_her_learnedB_r1_s2 +wandb_entity: amsks +wandb_group: fb_cube_her_learnedB_r1 +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__hernrm-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__hernrm-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ded6b98252382d2a7f150bf3d942c10dda6702ff --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__hernrm-s2/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_her_idnorm_r1_s2 +wandb_entity: amsks +wandb_group: fb_cube_her_idnorm_r1 +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: identity_norm +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__iqllb-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__iqllb-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ac4da9cb7bc159e5c14377fc31a65971888665c7 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__iqllb-s2/config.yaml @@ -0,0 +1,88 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_cube_iql_learnedB_r1_s2 +wandb_entity: amsks +wandb_group: fb_cube_iql_learnedB_r1 +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r0-base-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r0-base-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1ec763fab2004b402f9fa82aa325aa9a87b31f45 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r0-base-s2/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r0_base_s2 +wandb_entity: amsks +wandb_group: fb_bridge_r0_base +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: false +d_norm: false +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r1-decond-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r1-decond-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..88afb8367dd0954f8342adf8b3f95c1658648901 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r1-decond-s2/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r1_decond_s2 +wandb_entity: amsks +wandb_group: fb_bridge_r1_decond +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: true +d_norm: false +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r2-dnorm-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r2-dnorm-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..24942c420906b290c9ea28739cc56dc9d79b04a5 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r2-dnorm-s2/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r2_dnorm_s2 +wandb_entity: amsks +wandb_group: fb_bridge_r2_dnorm +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: false +d_norm: true +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r3-allon-s2/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r3-allon-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..beb9606a89e5541bc2ca2cf1fb6d5d5297255ba9 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s2__r3-allon-s2/config.yaml @@ -0,0 +1,100 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_bridge_r3_allon_s2 +wandb_entity: amsks +wandb_group: fb_bridge_r3_allon +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0003 +lr_b: 0.0003 +lr_actor: 0.0003 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +decouple_actor_z: false +critic_mode: iql +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +decondition_F: true +d_norm: true +critic_goal_mix: +- 0.5 +- 0.2 +- 0.3 +actor_goal_mix: +- 1.0 +- 0.0 +- 0.0 +iql_reward: goal_match +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s3__20260607-062100-fb-state-encmatch-s3/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s3__20260607-062100-fb-state-encmatch-s3/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1ec547d403bd70d2fd1c10b40551f460ae7344a2 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s3__20260607-062100-fb-state-encmatch-s3/config.yaml @@ -0,0 +1,80 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 3 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_encmatch_s3 +wandb_entity: amsks +wandb_group: fb_state_encmatch +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +rgb_encoder: + name: mlp + feature_dim: 256 + hidden_dim: 1100 + hidden_layers: 5 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s3__20260607-070225-fb-state-plain-s3/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s3__20260607-070225-fb-state-plain-s3/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6a348a81245a9a723896e8cc074f6db8104e25d9 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s3__20260607-070225-fb-state-plain-s3/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 3 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s3 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s4__20260607-062100-fb-state-encmatch-s4/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s4__20260607-062100-fb-state-encmatch-s4/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2853e2f47f5b4cd99e271a0bb476151ea1c96eda --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s4__20260607-062100-fb-state-encmatch-s4/config.yaml @@ -0,0 +1,80 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1000000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 4 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_encmatch_s4 +wandb_entity: amsks +wandb_group: fb_state_encmatch +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 256 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 0 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +rgb_encoder: + name: mlp + feature_dim: 256 + hidden_dim: 1100 + hidden_layers: 5 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s4__20260607-070225-fb-state-plain-s4/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s4__20260607-070225-fb-state-plain-s4/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a9d551dd3993fa0df42fbd0a899c0cec6d6b635f --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s4__20260607-070225-fb-state-plain-s4/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 4 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s4 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s5__20260607-070225-fb-state-plain-s5/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s5__20260607-070225-fb-state-plain-s5/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a2d99eb7428044ec9cc086a064269b57b82b31eb --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s5__20260607-070225-fb-state-plain-s5/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 5 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s5 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s6__20260607-070225-fb-state-plain-s6/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s6__20260607-070225-fb-state-plain-s6/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..703844e2c14f124ff14b83cbd0b5e8de0f57126d --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s6__20260607-070225-fb-state-plain-s6/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 6 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s6 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s7__20260607-070225-fb-state-plain-s7/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s7__20260607-070225-fb-state-plain-s7/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..57c62d5da8c319e052e42f8fed445d40856f886f --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s7__20260607-070225-fb-state-plain-s7/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 7 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s7 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s8__20260607-070225-fb-state-plain-s8/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s8__20260607-070225-fb-state-plain-s8/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7fe8d5a636594178aa3e0c44f88fe1e75ce2b09c --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s8__20260607-070225-fb-state-plain-s8/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 8 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s8 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s9__20260607-070225-fb-state-plain-s9/config.yaml b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s9__20260607-070225-fb-state-plain-s9/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..48a1ff5c5b528c35d5a0e1740ddaa72baa342847 --- /dev/null +++ b/checkpoints/torch/fb-flowbc/cube-single-play-v0-state__s9__20260607-070225-fb-state-plain-s9/config.yaml @@ -0,0 +1,78 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 1500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 9 +device: cuda +resume: true +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_plain_s9 +wandb_entity: amsks +wandb_group: fb_state_plain +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 diff --git a/checkpoints/torch/fb/cube-single-play-v0-state__s0__20260618-083115-fb-iql-ontraj-ddpgbc-s0/config.yaml b/checkpoints/torch/fb/cube-single-play-v0-state__s0__20260618-083115-fb-iql-ontraj-ddpgbc-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a07d757609ec5416823cb5f222fb151f2d4f85df --- /dev/null +++ b/checkpoints/torch/fb/cube-single-play-v0-state__s0__20260618-083115-fb-iql-ontraj-ddpgbc-s0/config.yaml @@ -0,0 +1,83 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_iql_ontraj_ddpgbc_s0 +wandb_entity: amsks +wandb_group: fb_iql_ontraj_ddpgbc +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 1 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: simple + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 diff --git a/checkpoints/torch/fb/cube-single-play-v0-state__s0__20260618-083115-fb-state-ddpgbc-s0/config.yaml b/checkpoints/torch/fb/cube-single-play-v0-state__s0__20260618-083115-fb-state-ddpgbc-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..456cfea6c42344c8ef02e3865e78a0a680d73e97 --- /dev/null +++ b/checkpoints/torch/fb/cube-single-play-v0-state__s0__20260618-083115-fb-state-ddpgbc-s0/config.yaml @@ -0,0 +1,83 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 0 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_ddpgbc_s0 +wandb_entity: amsks +wandb_group: fb_state_ddpgbc +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: perm +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: simple + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 diff --git a/checkpoints/torch/fb/cube-single-play-v0-state__s1__20260618-083115-fb-iql-ontraj-ddpgbc-s1/config.yaml b/checkpoints/torch/fb/cube-single-play-v0-state__s1__20260618-083115-fb-iql-ontraj-ddpgbc-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..afc624bb66373ff86cda6dd8e752a451e52ac777 --- /dev/null +++ b/checkpoints/torch/fb/cube-single-play-v0-state__s1__20260618-083115-fb-iql-ontraj-ddpgbc-s1/config.yaml @@ -0,0 +1,83 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_iql_ontraj_ddpgbc_s1 +wandb_entity: amsks +wandb_group: fb_iql_ontraj_ddpgbc +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 1 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: simple + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 diff --git a/checkpoints/torch/fb/cube-single-play-v0-state__s1__20260618-083115-fb-state-ddpgbc-s1/config.yaml b/checkpoints/torch/fb/cube-single-play-v0-state__s1__20260618-083115-fb-state-ddpgbc-s1/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..14230d0a070c129a1799068df3394fb7533f0b8f --- /dev/null +++ b/checkpoints/torch/fb/cube-single-play-v0-state__s1__20260618-083115-fb-state-ddpgbc-s1/config.yaml @@ -0,0 +1,83 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 1 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_ddpgbc_s1 +wandb_entity: amsks +wandb_group: fb_state_ddpgbc +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: perm +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: simple + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 diff --git a/checkpoints/torch/fb/cube-single-play-v0-state__s2__20260618-083115-fb-iql-ontraj-ddpgbc-s2/config.yaml b/checkpoints/torch/fb/cube-single-play-v0-state__s2__20260618-083115-fb-iql-ontraj-ddpgbc-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..94e7c357588b1297efdb47e6f366040fa95abbfb --- /dev/null +++ b/checkpoints/torch/fb/cube-single-play-v0-state__s2__20260618-083115-fb-iql-ontraj-ddpgbc-s2/config.yaml @@ -0,0 +1,83 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_iql_ontraj_ddpgbc_s2 +wandb_entity: amsks +wandb_group: fb_iql_ontraj_ddpgbc +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 1.0 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 1 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: traj +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: simple + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 diff --git a/checkpoints/torch/fb/cube-single-play-v0-state__s2__20260618-083115-fb-state-ddpgbc-s2/config.yaml b/checkpoints/torch/fb/cube-single-play-v0-state__s2__20260618-083115-fb-state-ddpgbc-s2/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d08681fca1f8212b8950374b7e4d8f9bfc20a11b --- /dev/null +++ b/checkpoints/torch/fb/cube-single-play-v0-state__s2__20260618-083115-fb-state-ddpgbc-s2/config.yaml @@ -0,0 +1,83 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 2 +device: cuda +resume: false +resume_from: null +force: false +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: fb_state_ddpgbc_s2 +wandb_entity: amsks +wandb_group: fb_state_ddpgbc +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: fb +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 1000 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +reweight_mode: coverage +iwr_c: 0.03 +iwr_sigma: 0.05 +iwr_eps: 0.1 +weight_diag: false +weight_z: false +train_goal_mode: perm +critic_mode: measure +iql_expectile: 0.9 +iql_thresh: 0.04 +onestep: false +goal_cond: false +fixed_b: none +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: simple + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 diff --git a/checkpoints/torch/rldp-flowbc/cube-single-play-v0-pixels__s4__20260604-052341-rldp-pixel-s4/config.yaml b/checkpoints/torch/rldp-flowbc/cube-single-play-v0-pixels__s4__20260604-052341-rldp-pixel-s4/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..06b3cf5506bcc6a0627e6dcd0153b01191e7ec71 --- /dev/null +++ b/checkpoints/torch/rldp-flowbc/cube-single-play-v0-pixels__s4__20260604-052341-rldp-pixel-s4/config.yaml @@ -0,0 +1,90 @@ +domain: cube-single-play-v0 +obs_type: pixels +data_path: /dev/shm/factored-fb/datasets_pixel +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 500000 +batch_size: 256 +eval_every: 50000 +log_every: 1000 +save_every: 200000 +seed: 4 +device: cuda +eval_n_episodes: 10 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: rldp_pixel_s4 +wandb_entity: amsks +wandb_group: rldp_pixel +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: rldp_flowbc +z_dim: 50 +L_dim: 50 +actor_encode_obs: false +discount: 0.99 +lr_f: 0.0001 +lr_b: 0.0001 +lr_actor: 0.0001 +weight_decay: 0.0 +clip_grad_norm: 0.0 +ortho_coef: 0.01 +train_goal_ratio: 0.5 +fb_pessimism_penalty: 0.0 +actor_pessimism_penalty: 0.0 +actor_std: 0.2 +stddev_clip: 0.3 +f_target_tau: 0.005 +b_target_tau: 0.005 +bc_coeff: 3.0 +q_loss_coef: 0.0 +reweight_alpha: 0.0 +reweight_clip: 10.0 +reweight_density_path: analysis/cube_density.npz +weight_diag: false +weight_z: false +onestep: false +amp: false +forward: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +backward: + hidden_dim: 512 + hidden_layers: 4 + norm: true +left_encoder: + hidden_dim: 512 + hidden_layers: 4 + norm: true +actor: + name: noise_conditioned_actor + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 +flow_steps: 10 +lr_actor_vf: 0.0003 +actor_vf: + hidden_dim: 512 + hidden_layers: 4 +horizon: 3 +predictor: + hidden_dim: 512 + hidden_layers: 2 + embedding_layers: 2 + num_parallel: 2 +frame_stack: 3 +obs_normalizer: + name: RGBNormalizerConfig +rgb_encoder: + name: drq + feature_dim: 256 +augmentator: + name: random_shifts + pad: 2 diff --git a/checkpoints/torch/tdmpc2/cube-single-play-v0-state__s0__tdmpc2-smoke10k-s0/config.yaml b/checkpoints/torch/tdmpc2/cube-single-play-v0-state__s0__tdmpc2-smoke10k-s0/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..003ce6126fc65902a574d8eaf89b984f8187cfeb --- /dev/null +++ b/checkpoints/torch/tdmpc2/cube-single-play-v0-state__s0__tdmpc2-smoke10k-s0/config.yaml @@ -0,0 +1,41 @@ +domain: cube-single-play-v0 +obs_type: state +data_path: /dev/shm/factored-fb/datasets +n_transitions: null +load_n_episodes: 1000 +num_train_steps: 10000 +batch_size: 256 +eval_every: 5000 +log_every: 200 +save_every: 200000 +seed: 0 +device: cuda +eval_n_episodes: 5 +eval_relabel_size: 10000 +eval_shift_reward: 1.0 +save_eval_videos: false +use_wandb: true +wandb_project: factored-fb +wandb_run_name: tdmpc2_smoke10k_s0 +wandb_entity: amsks +wandb_group: tdmpc2_smoke10k +wandb_tags: [] +wandb_save_code: true +save_dir: ${hydra:runtime.output_dir}/checkpoints +eval_videos_dir: ${hydra:runtime.output_dir}/eval_videos +agent: tdmpc2 +horizon: 3 +tdmpc2: + random_goal_ratio: 0.5 + success_threshold: 0.04 + latent_dim: 512 + mlp_dim: 512 + num_q: 5 + horizon: 3 + num_samples: 512 + num_elites: 64 + num_pi_trajs: 24 + iterations: 6 + episode_length: 200 +bc_coeff: 3.0 +lr_b: 0.0001