Upload 3 files

config.yaml

@@ -1,5 +1,5 @@
 seed: 0
-output_dir: output/03_05_2024_10_13_45_323746
+output_dir: output/08_05_2024_21_50_12_1563562
 domains: austin_sailor_dataset_converted_externally_to_rlds, stanford_hydra_dataset_converted_externally_to_rlds,
   austin_buds_dataset_converted_externally_to_rlds, austin_sirius_dataset_converted_externally_to_rlds,
   berkeley_mvp_converted_externally_to_rlds, berkeley_rpt_converted_externally_to_rlds,
@@ -13,15 +13,17 @@ domains: austin_sailor_dataset_converted_externally_to_rlds, stanford_hydra_data
   droid, uiuc_d3field, robo_set, qut_dexterous_manpulation, nyu_door_opening_surprising_effectiveness,
   nyu_franka_play_dataset_converted_externally_to_rlds, mimic_play, maniskill_dataset_converted_externally_to_rlds,
   columbia_cairlab_pusht_real, conq_hose_manipulation, dlr_edan_shared_control_converted_externally_to_rlds,
-  berkeley_gnm_sac_son, berkeley_autolab_ur5, aloha_mobile, agent_aware_affordances
-log_dir: output/03_05_2024_10_13_45_323746
+  berkeley_gnm_sac_son, berkeley_autolab_ur5, aloha_mobile, agent_aware_affordances,
+  human_image_hammer, human_image_spatula, epic_kitchen, drake_tooluse_image, pybullet_grasping_image,
+  mujoco_metaworld, mujoco_robomimic, isaac_arnold_image
+log_dir: output/08_05_2024_21_50_12_1563562
 debug_distributed: false
 wb_tag: default
-wb_cont_run: yel8colx
+wb_cont_run: 13h28mb6
 log_interval: 10
-script_name: run_resnet_30dataset_traj100000_embed1024_batch2048_800k
+script_name: run_resnet_30dataset_traj100000_embed_512_batch2048_800k_simhuman
 save_wb_checkpoint: true
-slurm_job_id: '27051970'
+slurm_job_id: '27327006'
 effective_total_epochs: 100
 effective_batch_size: 256
 epoch_size: 10
@@ -30,17 +32,17 @@ total_num_sample: 0
 rank: 0
 gpu: 0
 task_per_gpu: 1
-world_size: 64
+world_size: 32
 debug_submitit: false
 ngpus: 8
-nodes: 8
+nodes: 4
 timeout: 4320
 job_dir: logs/
 partition: learnlab
 use_volta32: true
 comment: ''
 resume: logs/
-dist_url: file:///checkpoint/xinleic/experiments/0171e314ce394a9d944eefc802a1747c_init
+dist_url: file:///checkpoint/xinleic/experiments/7b45bdd38afd4f85b05d324035d032b0_init
 dist_on_itp: false
 local_rank: 1
 distributed: true
@@ -81,7 +83,7 @@ dataset:
   continue_generate: false
 network:
   _target_: hpt_pretrain.models.policy.Policy
-  embed_dim: 1024
+  embed_dim: 512
   num_blocks: 16
   num_heads: 8
   use_modality_embedding: true
@@ -96,7 +98,7 @@ stem:
   modalities:
   - image
   - state
-  modality_embed_dim: 1024
+  modality_embed_dim: 512
   normalize_state: ${dataset.normalize_state}
   state_embedding_dim: 1
   image_encoder: ${dataset.image_encoder}
@@ -113,27 +115,27 @@ stem:
   image:
     _target_: hpt_pretrain.models.policy_stem.MLP
     input_dim: 512
-    output_dim: 1024
+    output_dim: 512
     widths:
     - 128
     num_of_copy: 1
   state:
     _target_: hpt_pretrain.models.policy_stem.MLP
-    input_dim: 8
-    output_dim: 1024
+    input_dim: 4
+    output_dim: 512
     widths:
     - 128
 head:
   _target_: hpt_pretrain.models.policy_head.MLP
-  input_dim: 1024
+  input_dim: 512
   tanh_end: true
-  output_dim: 48
+  output_dim: 56
   dropout: true
   widths:
   - 256
   - 128
 dataloader:
-  batch_size: 32
+  batch_size: 64
   num_workers: 1
   pin_memory: false
   persistent_workers: false
@@ -166,15 +168,15 @@ warmup_lr:
   step: 1000
 train:
   total_epochs: 30000
-  total_iters: 80000
+  total_iters: 800000
   epoch_iters: 1000
   validation_iters: 100
   use_accumulation: false
-  pretrained_dir: ''
+  pretrained_dir: '08_05_2024_21_50_12_1563562'
   max_validation_size: 10
   load_pretrain_trunk_only: false
   freeze_trunk: false
 lr_scheduler:
   _target_: torch.optim.lr_scheduler.CosineAnnealingLR
-  T_max: 80000
+  T_max: 800000
   eta_min: 1.0e-06

model.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1ff5ed46b9c235ec2ae547a124474334c7e5e86626ac199da06d866e8c99cad
-size 1624182299
+oid sha256:a0fab4ed6b34aec993b03752e461c5ef6f32d2107de50fcda6dc718c11b5f0d1
+size 699884363

trunk.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14d82dcf8287f1a2ad49480cfcdf01d190feff14a290e5e2292d909f7484580d
-size 806354858
+oid sha256:e1e96450b017ffca43e2acd7aba7d9744087b14065ca4e1752f5cd19f29dd30e
+size 201883562