made new model config hpt

egomimic/hydra_configs/data/mecka_all.yaml

@@ -0,0 +1,34 @@
+_target_: egomimic.pl_utils.pl_data_utils.MultiDataModuleWrapper
+
+
+train_datasets:
+  dataset1:
+    _target_: egomimic.rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
+    mode: train
+    local_mode: true
+    sample_percent: 0.1
+    filters:
+      lab: "mecka"
+    embodiment: "mecka_bimanual"
+
+valid_datasets:
+  dataset1:
+    _target_: egomimic.rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
+    mode: valid
+    local_mode: true
+    sample_percent: 0.1
+    filters:
+      lab: "mecka"
+    embodiment: "mecka_bimanual"
+
+train_dataloader_params:
+  dataset1:
+    batch_size: 32
+    num_workers: 8
+
+valid_dataloader_params:
+  dataset1:
+    batch_size: 32
+    num_workers: 8

egomimic/hydra_configs/data/mecka_cotrain.yaml

@@ -0,0 +1,50 @@
+_target_: egomimic.pl_utils.pl_data_utils.MultiDataModuleWrapper
+
+train_datasets:
+  # dataset1:
+  #   _target_: egomimic.rldb.utils.S3RLDBDataset
+  #   bucket_name: "rldb"
+  #   embodiment: "mecka_bimanual"
+  #   mode: train
+  #   filters: 
+  #     task: "potting_soil"
+  #   local_files_only: True
+  dataset2:
+    _target_: egomimic.rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
+    embodiment: "aria_right_arm"
+    mode: train
+    filters:
+      task: "object in container"
+      lab: "song"
+    local_files_only: True
+
+valid_datasets:
+  # dataset1:
+  #   _target_: egomimic.rldb.utils.S3RLDBDataset
+  #   bucket_name: "rldb"
+  #   embodiment: "mecka_bimanual"
+  #   mode: valid
+  #   filters: 
+  #     task: "potting_soil"
+  #   local_files_only: True
+  dataset2:
+    _target_: egomimic.rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
+    embodiment: "aria_right_arm"
+    mode: valid
+    filters: 
+      task: "object in container"
+      lab: "song"
+    local_files_only: True
+
+
+train_dataloader_params:
+  dataset1:
+    batch_size: 32
+    num_workers: 10
+
+valid_dataloader_params:
+  dataset1:
+    batch_size: 32
+    num_workers: 10

egomimic/hydra_configs/data/mecka_human_fold_clothes.yaml

@@ -0,0 +1,36 @@
+_target_: egomimic.pl_utils.pl_data_utils.MultiDataModuleWrapper
+
+train_datasets:
+  dataset1:
+    _target_: egomimic.rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
+    main_prefix: "mecka"
+    mode: train
+    embodiment: "mecka_bimanual"
+    temp_root: "/coc/flash7/scratch/egoverseS3Dataset"
+    filters: 
+      lab: "mecka"
+      task: "fold_clothes"
+    local_files_only: True
+
+valid_datasets:
+  dataset1:
+    _target_: egomimic.rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
+    mode: valid
+    embodiment: "mecka_bimanual"
+    temp_root: "/coc/flash7/scratch/egoverseS3Dataset"        
+    filters: 
+      lab: "mecka"
+      task: "fold_clothes"
+    local_files_only: True
+
+train_dataloader_params:
+  dataset1:
+    batch_size: 32
+    num_workers: 10
+
+valid_dataloader_params:
+  dataset1:
+    batch_size: 32
+    num_workers: 10

egomimic/hydra_configs/data/mecka_test.yaml

@@ -2,28 +2,41 @@ _target_: egomimic.pl_utils.pl_data_utils.MultiDataModuleWrapper
 
 train_datasets:
   dataset1:
-    _target_: rldb.utils.RLDBDataset
-    repo_id: "mecka_test"
+    _target_: rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
     mode: train
+    local_mode: True
     embodiment: "mecka_bimanual"
-    root: "/coc/flash7/acheluva3/EgoVerse/mecka_demo"
+    filters:
+      task: "fold_clothes"
     local_files_only: True
 
+
 valid_datasets:
   dataset1:
-    _target_: rldb.utils.RLDBDataset
-    repo_id: "mecka_test"
+    _target_: rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
     mode: valid
+    local_mode: True
     embodiment: "mecka_bimanual"
-    root: "/coc/flash7/acheluva3/EgoVerse/mecka_demo"
+    filters:
+      task: "fold_clothes"
     local_files_only: True
 
+
+
 train_dataloader_params:
   dataset1:
     batch_size: 32
     num_workers: 10
+  # dataset2:
+  #   batch_size: 32
+  #   num_workers: 10
 
 valid_dataloader_params:
   dataset1:
     batch_size: 32
-    num_workers: 10
+    num_workers: 10
+  # dataset2:
+  #   batch_size: 32
+  #   num_workers: 10

egomimic/hydra_configs/data/obj_cont_eva.yaml

@@ -0,0 +1,44 @@
+_target_: egomimic.pl_utils.pl_data_utils.MultiDataModuleWrapper
+use_tokenizer: false
+
+train_datasets:
+  dataset1:
+    _target_: egomimic.rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
+    temp_root: "/coc/cedarp-dxu345-0/datasets/egoverse"
+    mode: train
+    embodiment: "eva_right_arm"
+    filters: 
+      lab: "rl2"
+      task: "object in container"
+      robot_name: "eva_right_arm"
+      # objects:
+      #   - "green bowl"
+      #   - "caprisun"
+    local_files_only: True
+
+valid_datasets:
+  dataset1:
+    _target_: egomimic.rldb.utils.S3RLDBDataset
+    bucket_name: "rldb"
+    temp_root: "/coc/cedarp-dxu345-0/datasets/egoverse"
+    mode: valid
+    embodiment: "eva_right_arm"
+    filters: 
+      lab: "rl2"
+      task: "object in container"
+      robot_name: "eva_right_arm"
+      # objects:
+      #   - "green bowl"
+      #   - "caprisun"
+    local_files_only: True
+
+train_dataloader_params:
+  dataset1:
+    batch_size: 32
+    num_workers: 8
+
+valid_dataloader_params:
+  dataset1:
+    batch_size: 32
+    num_workers: 8

egomimic/hydra_configs/hydra/launcher/submitit_pace.yaml

@@ -5,13 +5,13 @@ _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher
 
 # Slurm configuration
 name: ${hydra.job.name}                             # Default job name
-partition: "gpu-h200"                             # Slurm partition
+# partition: "gpu-h200"                             # Slurm partition
 account: "gts-dxu345-rl2"                              # Slurm account
-cpus_per_task: 8                                   # Number of CPUs per task (max 4:1 CPU:GPU ratio)
+cpus_per_task: 4                                   # Number of CPUs per task (max 4:1 CPU:GPU ratio)
 nodes: ${launch_params.nodes}                       # Number of nodes
 tasks_per_node: ${launch_params.gpus_per_node}      # Use variable for tasks per node
 gres: "gpu:h200:${eval:'${launch_params.gpus_per_node} * ${launch_params.nodes}'}" # GPU type and count (h100 for H100 GPUs)
-qos: "short"                                        # Slurm QoS
+qos: "inferno"                                        # Slurm QoS
 mem_per_gpu: 250G
 timeout_min: 2880                                   # Timeout in minutes (48 hours)
 # exclude: "protocol, puma"     # Nodes to exclude

egomimic/hydra_configs/hydra/launcher/submitit_skynet.yaml

@@ -4,16 +4,23 @@ defaults:
 _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher
 
 name: ${hydra.job.name}
-partition: "rl2-lab"
-account: "rl2-lab"
+partition: "hoffman-lab"
+account: "hoffman-lab"
 # Override nodes via: hydra.launcher.nodes=<num_nodes>
-nodes: 1
 cpus_per_task: 15
-# Override via: hydra.launcher.tasks_per_node=<num_gpus>
-tasks_per_node: 4
+
+nodes: ${launch_params.nodes}                      
+tasks_per_node: ${launch_params.gpus_per_node}    
+gres: "gpu:a40:${eval:'${launch_params.gpus_per_node} * ${launch_params.nodes}'}"
+mem_per_gpu: 250G
+
 qos: "short"
-timeout_min: 2880
+timeout_min: 2880 # 48 hours
+
+# env:
+#   NCCL_TIMEOUT: 7200
+#   NCCL_DEBUG: INFO
+#   NCCL_ASYNC_ERROR_HANDLING: "1"
 
 additional_parameters:
-  requeue: true
-  gpus-per-node: "a40:1"
+  requeue: true

egomimic/hydra_configs/logger/wandb.yaml

@@ -7,7 +7,7 @@ wandb:
   offline: False
   id: "${name}_${description}_${now:%Y-%m-%d_%H-%M-%S}" # pass correct id to resume experiment!
   anonymous: null # enable anonymous logging
-  project: "everse_flagship"
+  project: "everse_mecka_hpt_ossification"
   log_model: False # upload lightning ckpts
   prefix: "" # a string to put at the beginning of metric keys
   entity: "rl2-group" # set to name of your wandb team