Upload folder using huggingface_hub

.gitattributes

@@ -49,3 +49,11 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 2025.12.17/04.26.39_train_llm_diffusion_unet_lowdim_parking-v0/wandb/run-20251217_042650-5vils759/run-5vils759.wandb filter=lfs diff=lfs merge=lfs -text
 2025.12.17/04.29.18_train_llm_diffusion_unet_lowdim_parking-v0/logs.json.txt filter=lfs diff=lfs merge=lfs -text
 2025.12.17/04.29.18_train_llm_diffusion_unet_lowdim_parking-v0/wandb/run-20251217_042930-etljn2u5/run-etljn2u5.wandb filter=lfs diff=lfs merge=lfs -text
+2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/run-dpa4txz7.wandb filter=lfs diff=lfs merge=lfs -text
+2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/run-wop5dm9n.wandb filter=lfs diff=lfs merge=lfs -text
+2025.12.18/04.43.54_train_llmbc_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2025.12.18/04.43.54_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044410-v5c3a4ag/run-v5c3a4ag.wandb filter=lfs diff=lfs merge=lfs -text
+2025.12.18/04.43.59_train_llmbc_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2025.12.18/04.43.59_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044411-dwnu9dgu/run-dwnu9dgu.wandb filter=lfs diff=lfs merge=lfs -text

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/.hydra/config.yaml

@@ -0,0 +1,163 @@
+name: train_llmbc_lowdim
+_target_: llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace
+obs_dim: ${task.obs_dim}
+action_dim: ${task.action_dim}
+task_name: ${task.name}
+exp_name: default
+model_name: ${llm.name}
+horizon: 1
+n_obs_steps: 1
+n_action_steps: 1
+n_latency_steps: 0
+past_action_visible: false
+llm_orig_expert_feedback: true
+llm_do_sample: false
+policy:
+  _target_: llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy
+  model:
+    _target_: llmbc.model.policy.policy_mlp.PolicyMLP
+    input_size: ${eval:'${n_obs_steps}*${obs_dim}'}
+    hidden_size:
+    - 256
+    - 256
+    output_size: ${eval:'${n_action_steps}*${action_dim}'}
+    activation: relu
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+  obs_dim: ${obs_dim}
+  action_dim: ${action_dim}
+  llm_discriminator:
+    _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+    task_id: ${task_name}
+    llm_translator:
+      _target_: llmbc.translator.llm_translator.LLMTranslator
+      cfg: ${llm}
+      obs_dim: ${task.obs_dim}
+      action_dim: ${task.action_dim}
+      horizon: ${horizon}
+      n_obs_steps: ${n_obs_steps}
+      n_action_steps: ${n_action_steps}
+  loss_bc_weight: 1.0
+  loss_llm_weight: 0.001
+  horizon: ${horizon}
+  n_obs_steps: ${n_obs_steps}
+  n_action_steps: ${n_action_steps}
+  normalize_llm_loss: true
+dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+val_dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 0.01
+  betas:
+  - 0.95
+  - 0.999
+  eps: 1.0e-08
+  weight_decay: 1.0e-06
+training:
+  device: cuda:0
+  seed: 43
+  debug: false
+  resume: false
+  lr_scheduler: cosine
+  lr_warmup_steps: 10
+  num_epochs: 20001
+  gradient_accumulate_every: 8
+  grad_norm_clip: 0.5
+  rollout_every: 100
+  checkpoint_every: 100
+  val_every: 1
+  sample_every: 100
+  sample_max_batch: 128
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-split-ablation
+  resume: true
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags:
+  - ${name}
+  - ${task_name}
+  - ${exp_name}
+  id: null
+  group: null
+checkpoint:
+  topk:
+    monitor_key: test_success_rate
+    mode: max
+    k: 5
+    format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+  save_last_ckpt: true
+  save_last_snapshot: false
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+task:
+  name: box-close-v2
+  obs_dim: 9
+  action_dim: 4
+  env_runner:
+    _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+    env_name: llf-metaworld-box-close-v2
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 30
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.9
+  dataset:
+    _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+    data_path: datasets/box-close-v2-split0.05.pt
+    data_path2: datasets/box-close-v2-split0.05.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: true
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: pretrained
+  finetune_mode: orig
+  checkpoint: data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+  max_length: 100
+  lora_config:
+    r: 32
+    lora_alpha: 64
+    lora_dropout: 0.05
+    bias: none
+    task_type: CAUSAL_LM
+  prompter:
+    _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/.hydra/hydra.yaml

@@ -0,0 +1,156 @@
+hydra:
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - training.seed=43
+    - policy.loss_llm_weight=1.0e-3
+  job:
+    name: train
+    chdir: null
+    override_dirname: policy.loss_llm_weight=1.0e-3,training.seed=43
+    id: ???
+    num: ???
+    config_name: llmbc_box-close-v2_pretrained_split0.05.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /root/workspace/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /root/workspace/LLM-BC/config/data_efficiency
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2
+    choices:
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/.hydra/overrides.yaml

@@ -0,0 +1,2 @@
+- training.seed=43
+- policy.loss_llm_weight=1.0e-3

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=3000-test_success_rate=0.720.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:673ec5fe8cf54fb74d71fc80acb33579737ae9a4bb123d4696e3d43c05b5c034
+size 864584

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=3500-test_success_rate=0.780.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c7658452945686e9f50361ad4305f66594ac1ce692658c0c03b8441ee04f1ba
+size 864584

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=4500-test_success_rate=0.780.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb4cd66395e1ae55e5c1179b88fd099fd5e80c91f1c5eb2b8ff136f8607092d9
+size 864584

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=5000-test_success_rate=0.780.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6f5a116439c1f64e48fcdb5f5e7e4885c42b246ca0f4bb18f845cf88cd427ec
+size 864584

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=6900-test_success_rate=0.860.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad4cb3264360accaa983e222418c180afc626fee3c408242f21e8213d3316436
+size 864584

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62b6e9fe67f92aae950175bd33e9224330d2b3642b149f04110fccc9b0b2aba4
+size 864584

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0bcd4d8517157ccb19a1fab7b30355bfce52e7e49ffb975249462565864de0f9
+size 55985396

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/train.log

@@ -0,0 +1,8 @@
+[2025-12-18 04:43:17,353][numexpr.utils][INFO] - Note: detected 128 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable.
+[2025-12-18 04:43:17,353][numexpr.utils][INFO] - Note: NumExpr detected 128 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2025-12-18 04:43:17,353][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2025-12-18 04:43:20,324][datasets][INFO] - PyTorch version 2.2.2 available.
+[2025-12-18 04:43:20,325][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2025-12-18 04:43:20,326][datasets][INFO] - JAX version 0.4.30 available.
+[2025-12-18 04:43:27,798][absl][INFO] - MUJOCO_GL=osmesa, attempting to import specified OpenGL backend.
+[2025-12-18 04:43:27,803][absl][INFO] - MuJoCo library version is: 2.3.7

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2025-12-18T04:43:28.833493537Z","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2025-12-18T04:43:28.833505356Z","level":"INFO","msg":"created symlink","path":"/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/logs/debug-core.log"}
+{"time":"2025-12-18T04:43:28.942203505Z","level":"INFO","msg":"created new stream","id":"dpa4txz7"}
+{"time":"2025-12-18T04:43:28.942234927Z","level":"INFO","msg":"stream: started","id":"dpa4txz7"}
+{"time":"2025-12-18T04:43:28.9422604Z","level":"INFO","msg":"sender: started","stream_id":"dpa4txz7"}
+{"time":"2025-12-18T04:43:28.942249251Z","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"dpa4txz7"}}
+{"time":"2025-12-18T04:43:28.942260477Z","level":"INFO","msg":"handler: started","stream_id":{"value":"dpa4txz7"}}
+{"time":"2025-12-18T04:43:30.17721562Z","level":"INFO","msg":"Starting system monitor"}
+{"time":"2025-12-19T12:01:09.038769606Z","level":"INFO","msg":"stream: closing","id":"dpa4txz7"}
+{"time":"2025-12-19T12:01:09.038800725Z","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2025-12-19T12:01:09.039292864Z","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2025-12-19T12:01:10.136222102Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-19T12:01:10.399575907Z","level":"INFO","msg":"handler: closed","stream_id":{"value":"dpa4txz7"}}
+{"time":"2025-12-19T12:01:10.399607638Z","level":"INFO","msg":"sender: closed","stream_id":"dpa4txz7"}
+{"time":"2025-12-19T12:01:10.399595523Z","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"dpa4txz7"}}
+{"time":"2025-12-19T12:01:10.399677461Z","level":"INFO","msg":"stream: closed","id":"dpa4txz7"}

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/debug.log

@@ -0,0 +1,27 @@
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Configure stats pid to 36909
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Loading settings from /root/.config/wandb/settings
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Loading settings from /root/workspace/LLM-BC/wandb/settings
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/root/workspace/LLM-BC/train.py', 'program': '/root/workspace/LLM-BC/./train.py'}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Applying login settings: {}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:_log_setup():533] Logging user logs to /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/logs/debug.log
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:_log_setup():534] Logging internal logs to /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/logs/debug-internal.log
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:init():619] calling init triggers
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:init():626] wandb.init called with sweep_config: {}
+config: {'name': 'train_llmbc_lowdim', '_target_': 'llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace', 'obs_dim': 9, 'action_dim': 4, 'task_name': 'box-close-v2', 'exp_name': 'default', 'model_name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'n_latency_steps': 0, 'past_action_visible': False, 'llm_orig_expert_feedback': True, 'llm_do_sample': False, 'policy': {'_target_': 'llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy', 'model': {'_target_': 'llmbc.model.policy.policy_mlp.PolicyMLP', 'input_size': 9, 'hidden_size': [256, 256], 'output_size': 4, 'activation': 'relu', 'n_obs_steps': 1, 'n_action_steps': 1}, 'obs_dim': 9, 'action_dim': 4, 'llm_discriminator': {'_target_': 'llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator', 'task_id': 'box-close-v2', 'llm_translator': {'_target_': 'llmbc.translator.llm_translator.LLMTranslator', 'cfg': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.18/04.43.16_HuggingFaceTB/SmolLM2-135M-Instruct'}}}, 'obs_dim': 9, 'action_dim': 4, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1}}, 'loss_bc_weight': 1.0, 'loss_llm_weight': 0.001, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'normalize_llm_loss': True}, 'dataloader': {'batch_size': 16, 'num_workers': 0, 'shuffle': True, 'pin_memory': False, 'persistent_workers': False}, 'val_dataloader': {'batch_size': 16, 'num_workers': 0, 'shuffle': True, 'pin_memory': False, 'persistent_workers': False}, 'optimizer': {'_target_': 'torch.optim.AdamW', 'lr': 0.01, 'betas': [0.95, 0.999], 'eps': 1e-08, 'weight_decay': 1e-06}, 'training': {'device': 'cuda:0', 'seed': 43, 'debug': False, 'resume': False, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 10, 'num_epochs': 20001, 'gradient_accumulate_every': 8, 'grad_norm_clip': 0.5, 'rollout_every': 100, 'checkpoint_every': 100, 'val_every': 1, 'sample_every': 100, 'sample_max_batch': 128, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-split-ablation', 'resume': True, 'mode': 'online', 'name': '2025.12.18-04.43.16_train_llmbc_lowdim_box-close-v2', 'tags': ['train_llmbc_lowdim', 'box-close-v2', 'default'], 'id': None, 'group': None}, 'checkpoint': {'topk': {'monitor_key': 'test_success_rate', 'mode': 'max', 'k': 5, 'format_str': 'epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt'}, 'save_last_ckpt': True, 'save_last_snapshot': False}, 'multi_run': {'run_dir': 'data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2', 'wandb_name_base': '2025.12.18-04.43.16_train_llmbc_lowdim_box-close-v2'}, 'task': {'name': 'box-close-v2', 'obs_dim': 9, 'action_dim': 4, 'env_runner': {'_target_': 'llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner', 'env_name': 'llf-metaworld-box-close-v2', 'n_train': 10, 'n_test': 50, 'n_envs': 10, 'max_steps': 30, 'n_obs_steps': 1, 'n_action_steps': 1, 'instruction_type': 'b', 'feedback_type': ['hp', 'hn', 'fp'], 'visual': False, 'discount': 0.9}, 'dataset': {'_target_': 'llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset', 'data_path': 'datasets/box-close-v2-split0.05.pt', 'data_path2': 'datasets/box-close-v2-split0.05.pt', 'horizon': 1, 'pad_before': 0, 'pad_after': 0, 'obs_eef_target': True, 'use_manual_normalizer': False, 'val_ratio': 0.1, 'dummy_normalizer': True}, 'instructor': {'_target_': 'llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor'}}, 'llm': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.18/04.43.16_HuggingFaceTB/SmolLM2-135M-Instruct'}}}}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:init():669] starting backend
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:init():673] sending inform_init request
+2025-12-18 04:43:28,832 INFO    MainThread:36909 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2025-12-18 04:43:28,832 INFO    MainThread:36909 [wandb_init.py:init():686] backend started and connected
+2025-12-18 04:43:28,839 INFO    MainThread:36909 [wandb_init.py:init():781] updated telemetry
+2025-12-18 04:43:28,867 INFO    MainThread:36909 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2025-12-18 04:43:30,174 INFO    MainThread:36909 [wandb_init.py:init():867] starting run threads in backend
+2025-12-18 04:43:30,515 INFO    MainThread:36909 [wandb_run.py:_console_start():2451] atexit reg
+2025-12-18 04:43:30,515 INFO    MainThread:36909 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2025-12-18 04:43:30,515 INFO    MainThread:36909 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2025-12-18 04:43:30,515 INFO    MainThread:36909 [wandb_run.py:_redirect():2389] Redirects installed.
+2025-12-18 04:43:30,516 INFO    MainThread:36909 [wandb_init.py:init():911] run started, returning control to user process
+2025-12-18 04:43:30,517 INFO    MainThread:36909 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2'}
+2025-12-19 12:01:09,038 WARNING MsgRouterThr:36909 [router.py:message_loop():75] message_loop has been closed

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/files/config.yaml

@@ -0,0 +1,271 @@
+_target_:
+    value: llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace
+_wandb:
+    value:
+        cli_version: 0.18.6
+        m: []
+        python_version: 3.9.25
+        t:
+            "1":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 11
+                - 12
+                - 41
+                - 49
+                - 50
+                - 51
+                - 53
+                - 55
+                - 71
+                - 83
+                - 95
+                - 98
+                - 100
+                - 105
+            "2":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 11
+                - 12
+                - 41
+                - 49
+                - 50
+                - 51
+                - 53
+                - 55
+                - 71
+                - 83
+                - 95
+                - 98
+                - 100
+                - 105
+            "3":
+                - 13
+                - 15
+                - 16
+                - 23
+                - 55
+                - 61
+            "4": 3.9.25
+            "5": 0.18.6
+            "6": 4.47.1
+            "8":
+                - 5
+            "12": 0.18.6
+            "13": linux-x86_64
+action_dim:
+    value: 4
+checkpoint:
+    value:
+        save_last_ckpt: true
+        save_last_snapshot: false
+        topk:
+            format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+            k: 5
+            mode: max
+            monitor_key: test_success_rate
+dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: true
+exp_name:
+    value: default
+horizon:
+    value: 1
+llm:
+    value:
+        causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+        checkpoint: data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+        config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+            run:
+                dir: data/outputs/2025.12.18/04.43.16_HuggingFaceTB/SmolLM2-135M-Instruct
+        llm_mode: pretrained
+        lora_config:
+            bias: none
+            lora_alpha: 64
+            lora_dropout: 0.05
+            r: 32
+            task_type: CAUSAL_LM
+        max_length: 100
+        model_name: SmolLM2-135M-Instruct
+        name: HuggingFaceTB/SmolLM2-135M-Instruct
+        prompter:
+            _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+            use_joint_mlp_projector: true
+        use_joint_mlp_projector: true
+        use_quantization: false
+llm_do_sample:
+    value: false
+llm_orig_expert_feedback:
+    value: true
+logging:
+    value:
+        group: null
+        id: null
+        mode: online
+        name: 2025.12.18-04.43.16_train_llmbc_lowdim_box-close-v2
+        project: box-close-v2-training-split-ablation
+        resume: true
+        tags:
+            - train_llmbc_lowdim
+            - box-close-v2
+            - default
+model_name:
+    value: HuggingFaceTB/SmolLM2-135M-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2
+        wandb_name_base: 2025.12.18-04.43.16_train_llmbc_lowdim_box-close-v2
+n_action_steps:
+    value: 1
+n_latency_steps:
+    value: 0
+n_obs_steps:
+    value: 1
+name:
+    value: train_llmbc_lowdim
+obs_dim:
+    value: 9
+optimizer:
+    value:
+        _target_: torch.optim.AdamW
+        betas:
+            - 0.95
+            - 0.999
+        eps: 1e-08
+        lr: 0.01
+        weight_decay: 1e-06
+output_dir:
+    value: /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2
+past_action_visible:
+    value: false
+policy:
+    value:
+        _target_: llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy
+        action_dim: 4
+        horizon: 1
+        llm_discriminator:
+            _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+            llm_translator:
+                _target_: llmbc.translator.llm_translator.LLMTranslator
+                action_dim: 4
+                cfg:
+                    causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+                    checkpoint: data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+                    config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+                        run:
+                            dir: data/outputs/2025.12.18/04.43.16_HuggingFaceTB/SmolLM2-135M-Instruct
+                    llm_mode: pretrained
+                    lora_config:
+                        bias: none
+                        lora_alpha: 64
+                        lora_dropout: 0.05
+                        r: 32
+                        task_type: CAUSAL_LM
+                    max_length: 100
+                    model_name: SmolLM2-135M-Instruct
+                    name: HuggingFaceTB/SmolLM2-135M-Instruct
+                    prompter:
+                        _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+                        use_joint_mlp_projector: true
+                    use_joint_mlp_projector: true
+                    use_quantization: false
+                horizon: 1
+                n_action_steps: 1
+                n_obs_steps: 1
+                obs_dim: 9
+            task_id: box-close-v2
+        loss_bc_weight: 1
+        loss_llm_weight: 0.001
+        model:
+            _target_: llmbc.model.policy.policy_mlp.PolicyMLP
+            activation: relu
+            hidden_size:
+                - 256
+                - 256
+            input_size: 9
+            n_action_steps: 1
+            n_obs_steps: 1
+            output_size: 4
+        n_action_steps: 1
+        n_obs_steps: 1
+        normalize_llm_loss: true
+        obs_dim: 9
+task:
+    value:
+        action_dim: 4
+        dataset:
+            _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+            data_path: datasets/box-close-v2-split0.05.pt
+            data_path2: datasets/box-close-v2-split0.05.pt
+            dummy_normalizer: true
+            horizon: 1
+            obs_eef_target: true
+            pad_after: 0
+            pad_before: 0
+            use_manual_normalizer: false
+            val_ratio: 0.1
+        env_runner:
+            _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+            discount: 0.9
+            env_name: llf-metaworld-box-close-v2
+            feedback_type:
+                - hp
+                - hn
+                - fp
+            instruction_type: b
+            max_steps: 30
+            n_action_steps: 1
+            n_envs: 10
+            n_obs_steps: 1
+            n_test: 50
+            n_train: 10
+            visual: false
+        instructor:
+            _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+        name: box-close-v2
+        obs_dim: 9
+task_name:
+    value: box-close-v2
+training:
+    value:
+        checkpoint_every: 100
+        debug: false
+        device: cuda:0
+        grad_norm_clip: 0.5
+        gradient_accumulate_every: 8
+        lr_scheduler: cosine
+        lr_warmup_steps: 10
+        max_train_steps: null
+        max_val_steps: null
+        num_epochs: 20001
+        resume: false
+        rollout_every: 100
+        sample_every: 100
+        sample_max_batch: 128
+        seed: 43
+        tqdm_interval_sec: 1
+        val_every: 1
+val_dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: true

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/files/output.log

@@ -0,0 +1,3 @@
+Eval MetaworldLowdimRunner 1/6:   0%|                                                                                               | 0/30 [00:00<?, ?it/s]/root/workspace/LLM-BC/llmbc/common/llfbench_util.py:39: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at ../torch/csrc/utils/tensor_new.cpp:275.)
+  obs = torch.tensor(obs, dtype=torch.float32).unsqueeze(dim=0).to(device)
+                                                                                                                                                           

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/files/wandb-metadata.json

@@ -0,0 +1,56 @@
+{
+  "os": "Linux-4.18.0-513.24.1.el8_9.x86_64-x86_64-with-glibc2.31",
+  "python": "3.9.25",
+  "startedAt": "2025-12-18T04:43:28.832802Z",
+  "args": [
+    "--config-path=./config/data_efficiency",
+    "--config-name",
+    "llmbc_box-close-v2_pretrained_split0.05.yaml",
+    "training.seed=43",
+    "policy.loss_llm_weight=1.0e-3"
+  ],
+  "program": "/root/workspace/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "615f426ea7020a632255502e2d2b273359968dce"
+  },
+  "email": "chris920325@gmail.com",
+  "root": "/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2",
+  "host": "llmbc-cuda-545d8cd8bc-9ltjf",
+  "username": "root",
+  "executable": "/root/miniconda3/envs/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 64,
+  "cpu_count_logical": 128,
+  "gpu": "NVIDIA H100 80GB HBM3",
+  "gpu_count": 2,
+  "disk": {
+    "/": {
+      "total": "23041585184768",
+      "used": "567559487488"
+    }
+  },
+  "memory": {
+    "total": "2163619737600"
+  },
+  "cpu": {
+    "count": 64,
+    "countLogical": 128
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    },
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    }
+  ],
+  "cudaVersion": "12.8"
+}

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/files/wandb-summary.json

@@ -0,0 +1 @@
+{"test/success_rate":0.58,"global_step":260012,"test/cumulative_reward":10.905790277616134,"train_loss":2.584472538886145e-07,"train_loss_llm":0.00016260147094726562,"lr":2.337295357257574e-11,"train/mean_score":4.898116436224846,"train/success_rate":0.5,"_wandb":{"runtime":112660},"grad_norm":5.9199439419899136e-05,"train_loss_bc":2.6557526666692866e-07,"val_loss":3.369854675838724e-05,"_step":260012,"train_action_mse_error":2.4320641500708007e-07,"epoch":20000,"_runtime":112660.205985026,"_timestamp":1.7661456673424504e+09,"test/mean_score":5.246551494127934,"train/cumulative_reward":12.676562884913375}

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2025-12-18T04:43:27.998019275Z","level":"INFO","msg":"started logging, with flags","port-filename":"/tmp/tmpr037_suq/port-36909.txt","pid":36909,"debug":false,"disable-analytics":false}
+{"time":"2025-12-18T04:43:27.998043571Z","level":"INFO","msg":"FeatureState","shutdownOnParentExitEnabled":false}
+{"time":"2025-12-18T04:43:28.003078214Z","level":"INFO","msg":"Will exit if parent process dies.","ppid":36909}
+{"time":"2025-12-18T04:43:28.00307352Z","level":"INFO","msg":"server is running","addr":{"IP":"127.0.0.1","Port":41381,"Zone":""}}
+{"time":"2025-12-18T04:43:28.192465473Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"127.0.0.1:50002"}
+{"time":"2025-12-18T04:43:28.833330458Z","level":"INFO","msg":"handleInformInit: received","streamId":"dpa4txz7","id":"127.0.0.1:50002"}
+{"time":"2025-12-18T04:43:28.942238099Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"dpa4txz7","id":"127.0.0.1:50002"}
+{"time":"2025-12-19T12:01:09.038689428Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"127.0.0.1:50002"}
+{"time":"2025-12-19T12:01:09.038778847Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2025-12-19T12:01:09.038765404Z","level":"INFO","msg":"connection: Close: initiating connection closure","id":"127.0.0.1:50002"}
+{"time":"2025-12-19T12:01:09.038860434Z","level":"INFO","msg":"connection: Close: connection successfully closed","id":"127.0.0.1:50002"}
+{"time":"2025-12-19T12:01:10.399715597Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"127.0.0.1:50002"}
+{"time":"2025-12-19T12:01:10.399726966Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"127.0.0.1:50002"}
+{"time":"2025-12-19T12:01:10.399732745Z","level":"INFO","msg":"server is closed"}

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/logs/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2025-12-18T04:43:28.833493537Z","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2025-12-18T04:43:28.833505356Z","level":"INFO","msg":"created symlink","path":"/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/logs/debug-core.log"}
+{"time":"2025-12-18T04:43:28.942203505Z","level":"INFO","msg":"created new stream","id":"dpa4txz7"}
+{"time":"2025-12-18T04:43:28.942234927Z","level":"INFO","msg":"stream: started","id":"dpa4txz7"}
+{"time":"2025-12-18T04:43:28.9422604Z","level":"INFO","msg":"sender: started","stream_id":"dpa4txz7"}
+{"time":"2025-12-18T04:43:28.942249251Z","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"dpa4txz7"}}
+{"time":"2025-12-18T04:43:28.942260477Z","level":"INFO","msg":"handler: started","stream_id":{"value":"dpa4txz7"}}
+{"time":"2025-12-18T04:43:30.17721562Z","level":"INFO","msg":"Starting system monitor"}
+{"time":"2025-12-19T12:01:09.038769606Z","level":"INFO","msg":"stream: closing","id":"dpa4txz7"}
+{"time":"2025-12-19T12:01:09.038800725Z","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2025-12-19T12:01:09.039292864Z","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2025-12-19T12:01:10.136222102Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-19T12:01:10.399575907Z","level":"INFO","msg":"handler: closed","stream_id":{"value":"dpa4txz7"}}
+{"time":"2025-12-19T12:01:10.399607638Z","level":"INFO","msg":"sender: closed","stream_id":"dpa4txz7"}
+{"time":"2025-12-19T12:01:10.399595523Z","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"dpa4txz7"}}
+{"time":"2025-12-19T12:01:10.399677461Z","level":"INFO","msg":"stream: closed","id":"dpa4txz7"}

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/logs/debug.log

@@ -0,0 +1,27 @@
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Configure stats pid to 36909
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Loading settings from /root/.config/wandb/settings
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Loading settings from /root/workspace/LLM-BC/wandb/settings
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/root/workspace/LLM-BC/train.py', 'program': '/root/workspace/LLM-BC/./train.py'}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_setup.py:_flush():79] Applying login settings: {}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:_log_setup():533] Logging user logs to /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/logs/debug.log
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:_log_setup():534] Logging internal logs to /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/logs/debug-internal.log
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:init():619] calling init triggers
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:init():626] wandb.init called with sweep_config: {}
+config: {'name': 'train_llmbc_lowdim', '_target_': 'llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace', 'obs_dim': 9, 'action_dim': 4, 'task_name': 'box-close-v2', 'exp_name': 'default', 'model_name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'n_latency_steps': 0, 'past_action_visible': False, 'llm_orig_expert_feedback': True, 'llm_do_sample': False, 'policy': {'_target_': 'llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy', 'model': {'_target_': 'llmbc.model.policy.policy_mlp.PolicyMLP', 'input_size': 9, 'hidden_size': [256, 256], 'output_size': 4, 'activation': 'relu', 'n_obs_steps': 1, 'n_action_steps': 1}, 'obs_dim': 9, 'action_dim': 4, 'llm_discriminator': {'_target_': 'llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator', 'task_id': 'box-close-v2', 'llm_translator': {'_target_': 'llmbc.translator.llm_translator.LLMTranslator', 'cfg': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.18/04.43.16_HuggingFaceTB/SmolLM2-135M-Instruct'}}}, 'obs_dim': 9, 'action_dim': 4, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1}}, 'loss_bc_weight': 1.0, 'loss_llm_weight': 0.001, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'normalize_llm_loss': True}, 'dataloader': {'batch_size': 16, 'num_workers': 0, 'shuffle': True, 'pin_memory': False, 'persistent_workers': False}, 'val_dataloader': {'batch_size': 16, 'num_workers': 0, 'shuffle': True, 'pin_memory': False, 'persistent_workers': False}, 'optimizer': {'_target_': 'torch.optim.AdamW', 'lr': 0.01, 'betas': [0.95, 0.999], 'eps': 1e-08, 'weight_decay': 1e-06}, 'training': {'device': 'cuda:0', 'seed': 43, 'debug': False, 'resume': False, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 10, 'num_epochs': 20001, 'gradient_accumulate_every': 8, 'grad_norm_clip': 0.5, 'rollout_every': 100, 'checkpoint_every': 100, 'val_every': 1, 'sample_every': 100, 'sample_max_batch': 128, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-split-ablation', 'resume': True, 'mode': 'online', 'name': '2025.12.18-04.43.16_train_llmbc_lowdim_box-close-v2', 'tags': ['train_llmbc_lowdim', 'box-close-v2', 'default'], 'id': None, 'group': None}, 'checkpoint': {'topk': {'monitor_key': 'test_success_rate', 'mode': 'max', 'k': 5, 'format_str': 'epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt'}, 'save_last_ckpt': True, 'save_last_snapshot': False}, 'multi_run': {'run_dir': 'data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2', 'wandb_name_base': '2025.12.18-04.43.16_train_llmbc_lowdim_box-close-v2'}, 'task': {'name': 'box-close-v2', 'obs_dim': 9, 'action_dim': 4, 'env_runner': {'_target_': 'llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner', 'env_name': 'llf-metaworld-box-close-v2', 'n_train': 10, 'n_test': 50, 'n_envs': 10, 'max_steps': 30, 'n_obs_steps': 1, 'n_action_steps': 1, 'instruction_type': 'b', 'feedback_type': ['hp', 'hn', 'fp'], 'visual': False, 'discount': 0.9}, 'dataset': {'_target_': 'llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset', 'data_path': 'datasets/box-close-v2-split0.05.pt', 'data_path2': 'datasets/box-close-v2-split0.05.pt', 'horizon': 1, 'pad_before': 0, 'pad_after': 0, 'obs_eef_target': True, 'use_manual_normalizer': False, 'val_ratio': 0.1, 'dummy_normalizer': True}, 'instructor': {'_target_': 'llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor'}}, 'llm': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.18/04.43.16_HuggingFaceTB/SmolLM2-135M-Instruct'}}}}
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:init():669] starting backend
+2025-12-18 04:43:28,831 INFO    MainThread:36909 [wandb_init.py:init():673] sending inform_init request
+2025-12-18 04:43:28,832 INFO    MainThread:36909 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2025-12-18 04:43:28,832 INFO    MainThread:36909 [wandb_init.py:init():686] backend started and connected
+2025-12-18 04:43:28,839 INFO    MainThread:36909 [wandb_init.py:init():781] updated telemetry
+2025-12-18 04:43:28,867 INFO    MainThread:36909 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2025-12-18 04:43:30,174 INFO    MainThread:36909 [wandb_init.py:init():867] starting run threads in backend
+2025-12-18 04:43:30,515 INFO    MainThread:36909 [wandb_run.py:_console_start():2451] atexit reg
+2025-12-18 04:43:30,515 INFO    MainThread:36909 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2025-12-18 04:43:30,515 INFO    MainThread:36909 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2025-12-18 04:43:30,515 INFO    MainThread:36909 [wandb_run.py:_redirect():2389] Redirects installed.
+2025-12-18 04:43:30,516 INFO    MainThread:36909 [wandb_init.py:init():911] run started, returning control to user process
+2025-12-18 04:43:30,517 INFO    MainThread:36909 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2'}
+2025-12-19 12:01:09,038 WARNING MsgRouterThr:36909 [router.py:message_loop():75] message_loop has been closed

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044328-dpa4txz7/run-dpa4txz7.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65cfda8c8eb8ba819ba6b747cc51dbd6b784d64b1efd7e1b70c9423ffdc26b94
+size 306687490

2025.12.18/04.43.16_train_llmbc_lowdim_box-close-v2/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "dpa4txz7"}

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/.hydra/config.yaml

@@ -0,0 +1,163 @@
+name: train_llmbc_lowdim
+_target_: llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace
+obs_dim: ${task.obs_dim}
+action_dim: ${task.action_dim}
+task_name: ${task.name}
+exp_name: default
+model_name: ${llm.name}
+horizon: 1
+n_obs_steps: 1
+n_action_steps: 1
+n_latency_steps: 0
+past_action_visible: false
+llm_orig_expert_feedback: true
+llm_do_sample: false
+policy:
+  _target_: llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy
+  model:
+    _target_: llmbc.model.policy.policy_mlp.PolicyMLP
+    input_size: ${eval:'${n_obs_steps}*${obs_dim}'}
+    hidden_size:
+    - 256
+    - 256
+    output_size: ${eval:'${n_action_steps}*${action_dim}'}
+    activation: relu
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+  obs_dim: ${obs_dim}
+  action_dim: ${action_dim}
+  llm_discriminator:
+    _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+    task_id: ${task_name}
+    llm_translator:
+      _target_: llmbc.translator.llm_translator.LLMTranslator
+      cfg: ${llm}
+      obs_dim: ${task.obs_dim}
+      action_dim: ${task.action_dim}
+      horizon: ${horizon}
+      n_obs_steps: ${n_obs_steps}
+      n_action_steps: ${n_action_steps}
+  loss_bc_weight: 1.0
+  loss_llm_weight: 0.001
+  horizon: ${horizon}
+  n_obs_steps: ${n_obs_steps}
+  n_action_steps: ${n_action_steps}
+  normalize_llm_loss: true
+dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+val_dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 0.01
+  betas:
+  - 0.95
+  - 0.999
+  eps: 1.0e-08
+  weight_decay: 1.0e-06
+training:
+  device: cuda:0
+  seed: 44
+  debug: false
+  resume: false
+  lr_scheduler: cosine
+  lr_warmup_steps: 10
+  num_epochs: 20001
+  gradient_accumulate_every: 8
+  grad_norm_clip: 0.5
+  rollout_every: 100
+  checkpoint_every: 100
+  val_every: 1
+  sample_every: 100
+  sample_max_batch: 128
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-split-ablation
+  resume: true
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags:
+  - ${name}
+  - ${task_name}
+  - ${exp_name}
+  id: null
+  group: null
+checkpoint:
+  topk:
+    monitor_key: test_success_rate
+    mode: max
+    k: 5
+    format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+  save_last_ckpt: true
+  save_last_snapshot: false
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+task:
+  name: box-close-v2
+  obs_dim: 9
+  action_dim: 4
+  env_runner:
+    _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+    env_name: llf-metaworld-box-close-v2
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 30
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.9
+  dataset:
+    _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+    data_path: datasets/box-close-v2-split0.05.pt
+    data_path2: datasets/box-close-v2-split0.05.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: true
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: pretrained
+  finetune_mode: orig
+  checkpoint: data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+  max_length: 100
+  lora_config:
+    r: 32
+    lora_alpha: 64
+    lora_dropout: 0.05
+    bias: none
+    task_type: CAUSAL_LM
+  prompter:
+    _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/.hydra/hydra.yaml

@@ -0,0 +1,156 @@
+hydra:
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - training.seed=44
+    - policy.loss_llm_weight=1.0e-3
+  job:
+    name: train
+    chdir: null
+    override_dirname: policy.loss_llm_weight=1.0e-3,training.seed=44
+    id: ???
+    num: ???
+    config_name: llmbc_box-close-v2_pretrained_split0.05.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /root/workspace/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /root/workspace/LLM-BC/config/data_efficiency
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2
+    choices:
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/.hydra/overrides.yaml

@@ -0,0 +1,2 @@
+- training.seed=44
+- policy.loss_llm_weight=1.0e-3

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=11400-test_success_rate=0.740.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b5c3ee18c7b79fefb1ac417260fe2a5175067ee584253de99a0f1ecb4dd72a6
+size 864584

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=13200-test_success_rate=0.740.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5dea0dbf8a632d59be3a11861fcc400aac4e011ae9c49b3712bcc4b562768867
+size 864584

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=13600-test_success_rate=0.740.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:323dee3c3b4819e4a82cbc0ace4a37eddb55234994661dd36492f41f31b47e35
+size 864584

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=14400-test_success_rate=0.740.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5bf58465a3a075b67c55a24d2839c424d4e1a69de0d200e8e7f754c20eafa30
+size 864584

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=14800-test_success_rate=0.740.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf18f54ed730a9ae9bba7c3d351217c781a723f4c8f2db8bfc743392b9129bd6
+size 864584

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc87666844f89077a042e2ac39031ba46d7f31dd3af41384b8d13f410892276d
+size 864584

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4938f8c5170ee2b1cdbad15080b89473e300469a39f1c7c13df5477341d304f0
+size 56065493

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/train.log

@@ -0,0 +1,8 @@
+[2025-12-18 04:43:25,466][numexpr.utils][INFO] - Note: detected 128 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable.
+[2025-12-18 04:43:25,466][numexpr.utils][INFO] - Note: NumExpr detected 128 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2025-12-18 04:43:25,466][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2025-12-18 04:43:28,417][datasets][INFO] - PyTorch version 2.2.2 available.
+[2025-12-18 04:43:28,418][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2025-12-18 04:43:28,418][datasets][INFO] - JAX version 0.4.30 available.
+[2025-12-18 04:43:36,029][absl][INFO] - MUJOCO_GL=osmesa, attempting to import specified OpenGL backend.
+[2025-12-18 04:43:36,035][absl][INFO] - MuJoCo library version is: 2.3.7

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2025-12-18T04:43:36.925727907Z","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2025-12-18T04:43:36.92573826Z","level":"INFO","msg":"created symlink","path":"/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/logs/debug-core.log"}
+{"time":"2025-12-18T04:43:37.137175165Z","level":"INFO","msg":"created new stream","id":"wop5dm9n"}
+{"time":"2025-12-18T04:43:37.137213573Z","level":"INFO","msg":"stream: started","id":"wop5dm9n"}
+{"time":"2025-12-18T04:43:37.137232723Z","level":"INFO","msg":"sender: started","stream_id":"wop5dm9n"}
+{"time":"2025-12-18T04:43:37.137228279Z","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"wop5dm9n"}}
+{"time":"2025-12-18T04:43:37.137235244Z","level":"INFO","msg":"handler: started","stream_id":{"value":"wop5dm9n"}}
+{"time":"2025-12-18T04:43:38.096908732Z","level":"INFO","msg":"Starting system monitor"}
+{"time":"2025-12-19T12:00:45.677705025Z","level":"INFO","msg":"stream: closing","id":"wop5dm9n"}
+{"time":"2025-12-19T12:00:45.677736117Z","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2025-12-19T12:00:45.678177689Z","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2025-12-19T12:00:46.863316656Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-19T12:00:47.371828406Z","level":"INFO","msg":"handler: closed","stream_id":{"value":"wop5dm9n"}}
+{"time":"2025-12-19T12:00:47.371863899Z","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"wop5dm9n"}}
+{"time":"2025-12-19T12:00:47.371890954Z","level":"INFO","msg":"sender: closed","stream_id":"wop5dm9n"}
+{"time":"2025-12-19T12:00:47.371930769Z","level":"INFO","msg":"stream: closed","id":"wop5dm9n"}

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/debug.log

@@ -0,0 +1,27 @@
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Configure stats pid to 37059
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Loading settings from /root/.config/wandb/settings
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Loading settings from /root/workspace/LLM-BC/wandb/settings
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/root/workspace/LLM-BC/train.py', 'program': '/root/workspace/LLM-BC/./train.py'}
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Applying login settings: {}
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_init.py:_log_setup():533] Logging user logs to /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/logs/debug.log
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_init.py:_log_setup():534] Logging internal logs to /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/logs/debug-internal.log
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_init.py:init():619] calling init triggers
+2025-12-18 04:43:36,923 INFO    MainThread:37059 [wandb_init.py:init():626] wandb.init called with sweep_config: {}
+config: {'name': 'train_llmbc_lowdim', '_target_': 'llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace', 'obs_dim': 9, 'action_dim': 4, 'task_name': 'box-close-v2', 'exp_name': 'default', 'model_name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'n_latency_steps': 0, 'past_action_visible': False, 'llm_orig_expert_feedback': True, 'llm_do_sample': False, 'policy': {'_target_': 'llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy', 'model': {'_target_': 'llmbc.model.policy.policy_mlp.PolicyMLP', 'input_size': 9, 'hidden_size': [256, 256], 'output_size': 4, 'activation': 'relu', 'n_obs_steps': 1, 'n_action_steps': 1}, 'obs_dim': 9, 'action_dim': 4, 'llm_discriminator': {'_target_': 'llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator', 'task_id': 'box-close-v2', 'llm_translator': {'_target_': 'llmbc.translator.llm_translator.LLMTranslator', 'cfg': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.18/04.43.24_HuggingFaceTB/SmolLM2-135M-Instruct'}}}, 'obs_dim': 9, 'action_dim': 4, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1}}, 'loss_bc_weight': 1.0, 'loss_llm_weight': 0.001, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'normalize_llm_loss': True}, 'dataloader': {'batch_size': 16, 'num_workers': 0, 'shuffle': True, 'pin_memory': False, 'persistent_workers': False}, 'val_dataloader': {'batch_size': 16, 'num_workers': 0, 'shuffle': True, 'pin_memory': False, 'persistent_workers': False}, 'optimizer': {'_target_': 'torch.optim.AdamW', 'lr': 0.01, 'betas': [0.95, 0.999], 'eps': 1e-08, 'weight_decay': 1e-06}, 'training': {'device': 'cuda:0', 'seed': 44, 'debug': False, 'resume': False, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 10, 'num_epochs': 20001, 'gradient_accumulate_every': 8, 'grad_norm_clip': 0.5, 'rollout_every': 100, 'checkpoint_every': 100, 'val_every': 1, 'sample_every': 100, 'sample_max_batch': 128, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-split-ablation', 'resume': True, 'mode': 'online', 'name': '2025.12.18-04.43.24_train_llmbc_lowdim_box-close-v2', 'tags': ['train_llmbc_lowdim', 'box-close-v2', 'default'], 'id': None, 'group': None}, 'checkpoint': {'topk': {'monitor_key': 'test_success_rate', 'mode': 'max', 'k': 5, 'format_str': 'epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt'}, 'save_last_ckpt': True, 'save_last_snapshot': False}, 'multi_run': {'run_dir': 'data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2', 'wandb_name_base': '2025.12.18-04.43.24_train_llmbc_lowdim_box-close-v2'}, 'task': {'name': 'box-close-v2', 'obs_dim': 9, 'action_dim': 4, 'env_runner': {'_target_': 'llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner', 'env_name': 'llf-metaworld-box-close-v2', 'n_train': 10, 'n_test': 50, 'n_envs': 10, 'max_steps': 30, 'n_obs_steps': 1, 'n_action_steps': 1, 'instruction_type': 'b', 'feedback_type': ['hp', 'hn', 'fp'], 'visual': False, 'discount': 0.9}, 'dataset': {'_target_': 'llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset', 'data_path': 'datasets/box-close-v2-split0.05.pt', 'data_path2': 'datasets/box-close-v2-split0.05.pt', 'horizon': 1, 'pad_before': 0, 'pad_after': 0, 'obs_eef_target': True, 'use_manual_normalizer': False, 'val_ratio': 0.1, 'dummy_normalizer': True}, 'instructor': {'_target_': 'llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor'}}, 'llm': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.18/04.43.24_HuggingFaceTB/SmolLM2-135M-Instruct'}}}}
+2025-12-18 04:43:36,923 INFO    MainThread:37059 [wandb_init.py:init():669] starting backend
+2025-12-18 04:43:36,923 INFO    MainThread:37059 [wandb_init.py:init():673] sending inform_init request
+2025-12-18 04:43:36,923 INFO    MainThread:37059 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2025-12-18 04:43:36,924 INFO    MainThread:37059 [wandb_init.py:init():686] backend started and connected
+2025-12-18 04:43:36,931 INFO    MainThread:37059 [wandb_init.py:init():781] updated telemetry
+2025-12-18 04:43:36,963 INFO    MainThread:37059 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2025-12-18 04:43:38,094 INFO    MainThread:37059 [wandb_init.py:init():867] starting run threads in backend
+2025-12-18 04:43:38,473 INFO    MainThread:37059 [wandb_run.py:_console_start():2451] atexit reg
+2025-12-18 04:43:38,473 INFO    MainThread:37059 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2025-12-18 04:43:38,473 INFO    MainThread:37059 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2025-12-18 04:43:38,473 INFO    MainThread:37059 [wandb_run.py:_redirect():2389] Redirects installed.
+2025-12-18 04:43:38,475 INFO    MainThread:37059 [wandb_init.py:init():911] run started, returning control to user process
+2025-12-18 04:43:38,475 INFO    MainThread:37059 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2'}
+2025-12-19 12:00:45,677 WARNING MsgRouterThr:37059 [router.py:message_loop():75] message_loop has been closed

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/files/config.yaml

@@ -0,0 +1,271 @@
+_target_:
+    value: llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace
+_wandb:
+    value:
+        cli_version: 0.18.6
+        m: []
+        python_version: 3.9.25
+        t:
+            "1":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 11
+                - 12
+                - 41
+                - 49
+                - 50
+                - 51
+                - 53
+                - 55
+                - 71
+                - 83
+                - 95
+                - 98
+                - 100
+                - 105
+            "2":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 11
+                - 12
+                - 41
+                - 49
+                - 50
+                - 51
+                - 53
+                - 55
+                - 71
+                - 83
+                - 95
+                - 98
+                - 100
+                - 105
+            "3":
+                - 13
+                - 15
+                - 16
+                - 23
+                - 55
+                - 61
+            "4": 3.9.25
+            "5": 0.18.6
+            "6": 4.47.1
+            "8":
+                - 5
+            "12": 0.18.6
+            "13": linux-x86_64
+action_dim:
+    value: 4
+checkpoint:
+    value:
+        save_last_ckpt: true
+        save_last_snapshot: false
+        topk:
+            format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+            k: 5
+            mode: max
+            monitor_key: test_success_rate
+dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: true
+exp_name:
+    value: default
+horizon:
+    value: 1
+llm:
+    value:
+        causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+        checkpoint: data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+        config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+            run:
+                dir: data/outputs/2025.12.18/04.43.24_HuggingFaceTB/SmolLM2-135M-Instruct
+        llm_mode: pretrained
+        lora_config:
+            bias: none
+            lora_alpha: 64
+            lora_dropout: 0.05
+            r: 32
+            task_type: CAUSAL_LM
+        max_length: 100
+        model_name: SmolLM2-135M-Instruct
+        name: HuggingFaceTB/SmolLM2-135M-Instruct
+        prompter:
+            _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+            use_joint_mlp_projector: true
+        use_joint_mlp_projector: true
+        use_quantization: false
+llm_do_sample:
+    value: false
+llm_orig_expert_feedback:
+    value: true
+logging:
+    value:
+        group: null
+        id: null
+        mode: online
+        name: 2025.12.18-04.43.24_train_llmbc_lowdim_box-close-v2
+        project: box-close-v2-training-split-ablation
+        resume: true
+        tags:
+            - train_llmbc_lowdim
+            - box-close-v2
+            - default
+model_name:
+    value: HuggingFaceTB/SmolLM2-135M-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2
+        wandb_name_base: 2025.12.18-04.43.24_train_llmbc_lowdim_box-close-v2
+n_action_steps:
+    value: 1
+n_latency_steps:
+    value: 0
+n_obs_steps:
+    value: 1
+name:
+    value: train_llmbc_lowdim
+obs_dim:
+    value: 9
+optimizer:
+    value:
+        _target_: torch.optim.AdamW
+        betas:
+            - 0.95
+            - 0.999
+        eps: 1e-08
+        lr: 0.01
+        weight_decay: 1e-06
+output_dir:
+    value: /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2
+past_action_visible:
+    value: false
+policy:
+    value:
+        _target_: llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy
+        action_dim: 4
+        horizon: 1
+        llm_discriminator:
+            _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+            llm_translator:
+                _target_: llmbc.translator.llm_translator.LLMTranslator
+                action_dim: 4
+                cfg:
+                    causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+                    checkpoint: data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+                    config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+                        run:
+                            dir: data/outputs/2025.12.18/04.43.24_HuggingFaceTB/SmolLM2-135M-Instruct
+                    llm_mode: pretrained
+                    lora_config:
+                        bias: none
+                        lora_alpha: 64
+                        lora_dropout: 0.05
+                        r: 32
+                        task_type: CAUSAL_LM
+                    max_length: 100
+                    model_name: SmolLM2-135M-Instruct
+                    name: HuggingFaceTB/SmolLM2-135M-Instruct
+                    prompter:
+                        _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+                        use_joint_mlp_projector: true
+                    use_joint_mlp_projector: true
+                    use_quantization: false
+                horizon: 1
+                n_action_steps: 1
+                n_obs_steps: 1
+                obs_dim: 9
+            task_id: box-close-v2
+        loss_bc_weight: 1
+        loss_llm_weight: 0.001
+        model:
+            _target_: llmbc.model.policy.policy_mlp.PolicyMLP
+            activation: relu
+            hidden_size:
+                - 256
+                - 256
+            input_size: 9
+            n_action_steps: 1
+            n_obs_steps: 1
+            output_size: 4
+        n_action_steps: 1
+        n_obs_steps: 1
+        normalize_llm_loss: true
+        obs_dim: 9
+task:
+    value:
+        action_dim: 4
+        dataset:
+            _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+            data_path: datasets/box-close-v2-split0.05.pt
+            data_path2: datasets/box-close-v2-split0.05.pt
+            dummy_normalizer: true
+            horizon: 1
+            obs_eef_target: true
+            pad_after: 0
+            pad_before: 0
+            use_manual_normalizer: false
+            val_ratio: 0.1
+        env_runner:
+            _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+            discount: 0.9
+            env_name: llf-metaworld-box-close-v2
+            feedback_type:
+                - hp
+                - hn
+                - fp
+            instruction_type: b
+            max_steps: 30
+            n_action_steps: 1
+            n_envs: 10
+            n_obs_steps: 1
+            n_test: 50
+            n_train: 10
+            visual: false
+        instructor:
+            _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+        name: box-close-v2
+        obs_dim: 9
+task_name:
+    value: box-close-v2
+training:
+    value:
+        checkpoint_every: 100
+        debug: false
+        device: cuda:0
+        grad_norm_clip: 0.5
+        gradient_accumulate_every: 8
+        lr_scheduler: cosine
+        lr_warmup_steps: 10
+        max_train_steps: null
+        max_val_steps: null
+        num_epochs: 20001
+        resume: false
+        rollout_every: 100
+        sample_every: 100
+        sample_max_batch: 128
+        seed: 44
+        tqdm_interval_sec: 1
+        val_every: 1
+val_dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: true

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/files/output.log

@@ -0,0 +1,3 @@
+Eval MetaworldLowdimRunner 1/6:   0%|                                                                                               | 0/30 [00:00<?, ?it/s]/root/workspace/LLM-BC/llmbc/common/llfbench_util.py:39: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at ../torch/csrc/utils/tensor_new.cpp:275.)
+  obs = torch.tensor(obs, dtype=torch.float32).unsqueeze(dim=0).to(device)
+                                                                                                                                                           

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/files/wandb-metadata.json

@@ -0,0 +1,56 @@
+{
+  "os": "Linux-4.18.0-513.24.1.el8_9.x86_64-x86_64-with-glibc2.31",
+  "python": "3.9.25",
+  "startedAt": "2025-12-18T04:43:36.924319Z",
+  "args": [
+    "--config-path=./config/data_efficiency",
+    "--config-name",
+    "llmbc_box-close-v2_pretrained_split0.05.yaml",
+    "training.seed=44",
+    "policy.loss_llm_weight=1.0e-3"
+  ],
+  "program": "/root/workspace/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "615f426ea7020a632255502e2d2b273359968dce"
+  },
+  "email": "chris920325@gmail.com",
+  "root": "/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2",
+  "host": "llmbc-cuda-545d8cd8bc-9ltjf",
+  "username": "root",
+  "executable": "/root/miniconda3/envs/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 64,
+  "cpu_count_logical": 128,
+  "gpu": "NVIDIA H100 80GB HBM3",
+  "gpu_count": 2,
+  "disk": {
+    "/": {
+      "total": "23041585184768",
+      "used": "567559561216"
+    }
+  },
+  "memory": {
+    "total": "2163619737600"
+  },
+  "cpu": {
+    "count": 64,
+    "countLogical": 128
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    },
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    }
+  ],
+  "cudaVersion": "12.8"
+}

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_wandb":{"runtime":112628},"train_loss":2.4937074490830685e-07,"test/cumulative_reward":11.212119010439606,"grad_norm":3.27207162627019e-05,"_step":260012,"lr":2.337295357257574e-11,"train/success_rate":0.8,"train_loss_llm":2.1219253540039062e-05,"test/mean_score":5.510511810154676,"train/mean_score":7.082076881683081,"epoch":20000,"val_loss":2.5340521460748278e-05,"global_step":260012,"train/cumulative_reward":14.487907827312075,"_runtime":112628.753404656,"train_loss_bc":2.329571913151085e-07,"_timestamp":1.7661456439568613e+09,"test/success_rate":0.6,"train_action_mse_error":2.9938050261080207e-07}

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2025-12-18T04:43:36.244891686Z","level":"INFO","msg":"started logging, with flags","port-filename":"/tmp/tmperpi1ugw/port-37059.txt","pid":37059,"debug":false,"disable-analytics":false}
+{"time":"2025-12-18T04:43:36.244922397Z","level":"INFO","msg":"FeatureState","shutdownOnParentExitEnabled":false}
+{"time":"2025-12-18T04:43:36.246089731Z","level":"INFO","msg":"Will exit if parent process dies.","ppid":37059}
+{"time":"2025-12-18T04:43:36.24609079Z","level":"INFO","msg":"server is running","addr":{"IP":"127.0.0.1","Port":33233,"Zone":""}}
+{"time":"2025-12-18T04:43:36.439727768Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"127.0.0.1:34488"}
+{"time":"2025-12-18T04:43:36.925578675Z","level":"INFO","msg":"handleInformInit: received","streamId":"wop5dm9n","id":"127.0.0.1:34488"}
+{"time":"2025-12-18T04:43:37.137218722Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"wop5dm9n","id":"127.0.0.1:34488"}
+{"time":"2025-12-19T12:00:45.67764808Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"127.0.0.1:34488"}
+{"time":"2025-12-19T12:00:45.677700887Z","level":"INFO","msg":"connection: Close: initiating connection closure","id":"127.0.0.1:34488"}
+{"time":"2025-12-19T12:00:45.677712049Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2025-12-19T12:00:45.677751454Z","level":"INFO","msg":"connection: Close: connection successfully closed","id":"127.0.0.1:34488"}
+{"time":"2025-12-19T12:00:47.371978307Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"127.0.0.1:34488"}
+{"time":"2025-12-19T12:00:47.371996502Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"127.0.0.1:34488"}
+{"time":"2025-12-19T12:00:47.372006394Z","level":"INFO","msg":"server is closed"}

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/logs/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2025-12-18T04:43:36.925727907Z","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2025-12-18T04:43:36.92573826Z","level":"INFO","msg":"created symlink","path":"/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/logs/debug-core.log"}
+{"time":"2025-12-18T04:43:37.137175165Z","level":"INFO","msg":"created new stream","id":"wop5dm9n"}
+{"time":"2025-12-18T04:43:37.137213573Z","level":"INFO","msg":"stream: started","id":"wop5dm9n"}
+{"time":"2025-12-18T04:43:37.137232723Z","level":"INFO","msg":"sender: started","stream_id":"wop5dm9n"}
+{"time":"2025-12-18T04:43:37.137228279Z","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"wop5dm9n"}}
+{"time":"2025-12-18T04:43:37.137235244Z","level":"INFO","msg":"handler: started","stream_id":{"value":"wop5dm9n"}}
+{"time":"2025-12-18T04:43:38.096908732Z","level":"INFO","msg":"Starting system monitor"}
+{"time":"2025-12-19T12:00:45.677705025Z","level":"INFO","msg":"stream: closing","id":"wop5dm9n"}
+{"time":"2025-12-19T12:00:45.677736117Z","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2025-12-19T12:00:45.678177689Z","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2025-12-19T12:00:46.863316656Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-19T12:00:47.371828406Z","level":"INFO","msg":"handler: closed","stream_id":{"value":"wop5dm9n"}}
+{"time":"2025-12-19T12:00:47.371863899Z","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"wop5dm9n"}}
+{"time":"2025-12-19T12:00:47.371890954Z","level":"INFO","msg":"sender: closed","stream_id":"wop5dm9n"}
+{"time":"2025-12-19T12:00:47.371930769Z","level":"INFO","msg":"stream: closed","id":"wop5dm9n"}

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/logs/debug.log

@@ -0,0 +1,27 @@
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Configure stats pid to 37059
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Loading settings from /root/.config/wandb/settings
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Loading settings from /root/workspace/LLM-BC/wandb/settings
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/root/workspace/LLM-BC/train.py', 'program': '/root/workspace/LLM-BC/./train.py'}
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_setup.py:_flush():79] Applying login settings: {}
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_init.py:_log_setup():533] Logging user logs to /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/logs/debug.log
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_init.py:_log_setup():534] Logging internal logs to /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/logs/debug-internal.log
+2025-12-18 04:43:36,922 INFO    MainThread:37059 [wandb_init.py:init():619] calling init triggers
+2025-12-18 04:43:36,923 INFO    MainThread:37059 [wandb_init.py:init():626] wandb.init called with sweep_config: {}
+config: {'name': 'train_llmbc_lowdim', '_target_': 'llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace', 'obs_dim': 9, 'action_dim': 4, 'task_name': 'box-close-v2', 'exp_name': 'default', 'model_name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'n_latency_steps': 0, 'past_action_visible': False, 'llm_orig_expert_feedback': True, 'llm_do_sample': False, 'policy': {'_target_': 'llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy', 'model': {'_target_': 'llmbc.model.policy.policy_mlp.PolicyMLP', 'input_size': 9, 'hidden_size': [256, 256], 'output_size': 4, 'activation': 'relu', 'n_obs_steps': 1, 'n_action_steps': 1}, 'obs_dim': 9, 'action_dim': 4, 'llm_discriminator': {'_target_': 'llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator', 'task_id': 'box-close-v2', 'llm_translator': {'_target_': 'llmbc.translator.llm_translator.LLMTranslator', 'cfg': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.18/04.43.24_HuggingFaceTB/SmolLM2-135M-Instruct'}}}, 'obs_dim': 9, 'action_dim': 4, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1}}, 'loss_bc_weight': 1.0, 'loss_llm_weight': 0.001, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'normalize_llm_loss': True}, 'dataloader': {'batch_size': 16, 'num_workers': 0, 'shuffle': True, 'pin_memory': False, 'persistent_workers': False}, 'val_dataloader': {'batch_size': 16, 'num_workers': 0, 'shuffle': True, 'pin_memory': False, 'persistent_workers': False}, 'optimizer': {'_target_': 'torch.optim.AdamW', 'lr': 0.01, 'betas': [0.95, 0.999], 'eps': 1e-08, 'weight_decay': 1e-06}, 'training': {'device': 'cuda:0', 'seed': 44, 'debug': False, 'resume': False, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 10, 'num_epochs': 20001, 'gradient_accumulate_every': 8, 'grad_norm_clip': 0.5, 'rollout_every': 100, 'checkpoint_every': 100, 'val_every': 1, 'sample_every': 100, 'sample_max_batch': 128, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-split-ablation', 'resume': True, 'mode': 'online', 'name': '2025.12.18-04.43.24_train_llmbc_lowdim_box-close-v2', 'tags': ['train_llmbc_lowdim', 'box-close-v2', 'default'], 'id': None, 'group': None}, 'checkpoint': {'topk': {'monitor_key': 'test_success_rate', 'mode': 'max', 'k': 5, 'format_str': 'epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt'}, 'save_last_ckpt': True, 'save_last_snapshot': False}, 'multi_run': {'run_dir': 'data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2', 'wandb_name_base': '2025.12.18-04.43.24_train_llmbc_lowdim_box-close-v2'}, 'task': {'name': 'box-close-v2', 'obs_dim': 9, 'action_dim': 4, 'env_runner': {'_target_': 'llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner', 'env_name': 'llf-metaworld-box-close-v2', 'n_train': 10, 'n_test': 50, 'n_envs': 10, 'max_steps': 30, 'n_obs_steps': 1, 'n_action_steps': 1, 'instruction_type': 'b', 'feedback_type': ['hp', 'hn', 'fp'], 'visual': False, 'discount': 0.9}, 'dataset': {'_target_': 'llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset', 'data_path': 'datasets/box-close-v2-split0.05.pt', 'data_path2': 'datasets/box-close-v2-split0.05.pt', 'horizon': 1, 'pad_before': 0, 'pad_after': 0, 'obs_eef_target': True, 'use_manual_normalizer': False, 'val_ratio': 0.1, 'dummy_normalizer': True}, 'instructor': {'_target_': 'llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor'}}, 'llm': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.18/04.43.24_HuggingFaceTB/SmolLM2-135M-Instruct'}}}}
+2025-12-18 04:43:36,923 INFO    MainThread:37059 [wandb_init.py:init():669] starting backend
+2025-12-18 04:43:36,923 INFO    MainThread:37059 [wandb_init.py:init():673] sending inform_init request
+2025-12-18 04:43:36,923 INFO    MainThread:37059 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2025-12-18 04:43:36,924 INFO    MainThread:37059 [wandb_init.py:init():686] backend started and connected
+2025-12-18 04:43:36,931 INFO    MainThread:37059 [wandb_init.py:init():781] updated telemetry
+2025-12-18 04:43:36,963 INFO    MainThread:37059 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2025-12-18 04:43:38,094 INFO    MainThread:37059 [wandb_init.py:init():867] starting run threads in backend
+2025-12-18 04:43:38,473 INFO    MainThread:37059 [wandb_run.py:_console_start():2451] atexit reg
+2025-12-18 04:43:38,473 INFO    MainThread:37059 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2025-12-18 04:43:38,473 INFO    MainThread:37059 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2025-12-18 04:43:38,473 INFO    MainThread:37059 [wandb_run.py:_redirect():2389] Redirects installed.
+2025-12-18 04:43:38,475 INFO    MainThread:37059 [wandb_init.py:init():911] run started, returning control to user process
+2025-12-18 04:43:38,475 INFO    MainThread:37059 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2'}
+2025-12-19 12:00:45,677 WARNING MsgRouterThr:37059 [router.py:message_loop():75] message_loop has been closed

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/run-20251218_044336-wop5dm9n/run-wop5dm9n.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e29913bbfabcda09c18489ea82538ed60c1bbc47ff5972cd4d3909bf27bfbff4
+size 306855502

2025.12.18/04.43.24_train_llmbc_lowdim_box-close-v2/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "wop5dm9n"}

2025.12.18/04.43.35_train_llmbc_lowdim_box-close-v2/.hydra/config.yaml

@@ -0,0 +1,163 @@
+name: train_llmbc_lowdim
+_target_: llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace
+obs_dim: ${task.obs_dim}
+action_dim: ${task.action_dim}
+task_name: ${task.name}
+exp_name: default
+model_name: ${llm.name}
+horizon: 1
+n_obs_steps: 1
+n_action_steps: 1
+n_latency_steps: 0
+past_action_visible: false
+llm_orig_expert_feedback: true
+llm_do_sample: false
+policy:
+  _target_: llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy
+  model:
+    _target_: llmbc.model.policy.policy_mlp.PolicyMLP
+    input_size: ${eval:'${n_obs_steps}*${obs_dim}'}
+    hidden_size:
+    - 256
+    - 256
+    output_size: ${eval:'${n_action_steps}*${action_dim}'}
+    activation: relu
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+  obs_dim: ${obs_dim}
+  action_dim: ${action_dim}
+  llm_discriminator:
+    _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+    task_id: ${task_name}
+    llm_translator:
+      _target_: llmbc.translator.llm_translator.LLMTranslator
+      cfg: ${llm}
+      obs_dim: ${task.obs_dim}
+      action_dim: ${task.action_dim}
+      horizon: ${horizon}
+      n_obs_steps: ${n_obs_steps}
+      n_action_steps: ${n_action_steps}
+  loss_bc_weight: 1.0
+  loss_llm_weight: 0.001
+  horizon: ${horizon}
+  n_obs_steps: ${n_obs_steps}
+  n_action_steps: ${n_action_steps}
+  normalize_llm_loss: true
+dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+val_dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 0.01
+  betas:
+  - 0.95
+  - 0.999
+  eps: 1.0e-08
+  weight_decay: 1.0e-06
+training:
+  device: cuda:0
+  seed: 45
+  debug: false
+  resume: false
+  lr_scheduler: cosine
+  lr_warmup_steps: 10
+  num_epochs: 20001
+  gradient_accumulate_every: 8
+  grad_norm_clip: 0.5
+  rollout_every: 100
+  checkpoint_every: 100
+  val_every: 1
+  sample_every: 100
+  sample_max_batch: 128
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-split-ablation
+  resume: true
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags:
+  - ${name}
+  - ${task_name}
+  - ${exp_name}
+  id: null
+  group: null
+checkpoint:
+  topk:
+    monitor_key: test_success_rate
+    mode: max
+    k: 5
+    format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+  save_last_ckpt: true
+  save_last_snapshot: false
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+task:
+  name: box-close-v2
+  obs_dim: 9
+  action_dim: 4
+  env_runner:
+    _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+    env_name: llf-metaworld-box-close-v2
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 30
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.9
+  dataset:
+    _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+    data_path: datasets/box-close-v2-split0.05.pt
+    data_path2: datasets/box-close-v2-split0.05.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: true
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: pretrained
+  finetune_mode: orig
+  checkpoint: data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+  max_length: 100
+  lora_config:
+    r: 32
+    lora_alpha: 64
+    lora_dropout: 0.05
+    bias: none
+    task_type: CAUSAL_LM
+  prompter:
+    _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}

2025.12.18/04.43.35_train_llmbc_lowdim_box-close-v2/.hydra/hydra.yaml

@@ -0,0 +1,156 @@
+hydra:
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - training.seed=45
+    - policy.loss_llm_weight=1.0e-3
+  job:
+    name: train
+    chdir: null
+    override_dirname: policy.loss_llm_weight=1.0e-3,training.seed=45
+    id: ???
+    num: ???
+    config_name: llmbc_box-close-v2_pretrained_split0.05.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /root/workspace/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /root/workspace/LLM-BC/config/data_efficiency
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /root/workspace/LLM-BC/data/outputs/2025.12.18/04.43.35_train_llmbc_lowdim_box-close-v2
+    choices:
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

2025.12.18/04.43.35_train_llmbc_lowdim_box-close-v2/.hydra/overrides.yaml

@@ -0,0 +1,2 @@
+- training.seed=45
+- policy.loss_llm_weight=1.0e-3

2025.12.18/04.43.35_train_llmbc_lowdim_box-close-v2/train.log

@@ -0,0 +1,6 @@
+[2025-12-18 04:43:36,385][numexpr.utils][INFO] - Note: detected 128 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable.
+[2025-12-18 04:43:36,386][numexpr.utils][INFO] - Note: NumExpr detected 128 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2025-12-18 04:43:36,386][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2025-12-18 04:43:40,556][datasets][INFO] - PyTorch version 2.2.2 available.
+[2025-12-18 04:43:40,557][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2025-12-18 04:43:40,558][datasets][INFO] - JAX version 0.4.30 available.

2025.12.18/04.43.54_train_llmbc_lowdim_box-close-v2/.hydra/config.yaml

@@ -0,0 +1,163 @@
+name: train_llmbc_lowdim
+_target_: llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace
+obs_dim: ${task.obs_dim}
+action_dim: ${task.action_dim}
+task_name: ${task.name}
+exp_name: default
+model_name: ${llm.name}
+horizon: 1
+n_obs_steps: 1
+n_action_steps: 1
+n_latency_steps: 0
+past_action_visible: false
+llm_orig_expert_feedback: true
+llm_do_sample: false
+policy:
+  _target_: llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy
+  model:
+    _target_: llmbc.model.policy.policy_mlp.PolicyMLP
+    input_size: ${eval:'${n_obs_steps}*${obs_dim}'}
+    hidden_size:
+    - 256
+    - 256
+    output_size: ${eval:'${n_action_steps}*${action_dim}'}
+    activation: relu
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+  obs_dim: ${obs_dim}
+  action_dim: ${action_dim}
+  llm_discriminator:
+    _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+    task_id: ${task_name}
+    llm_translator:
+      _target_: llmbc.translator.llm_translator.LLMTranslator
+      cfg: ${llm}
+      obs_dim: ${task.obs_dim}
+      action_dim: ${task.action_dim}
+      horizon: ${horizon}
+      n_obs_steps: ${n_obs_steps}
+      n_action_steps: ${n_action_steps}
+  loss_bc_weight: 1.0
+  loss_llm_weight: 0.001
+  horizon: ${horizon}
+  n_obs_steps: ${n_obs_steps}
+  n_action_steps: ${n_action_steps}
+  normalize_llm_loss: true
+dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+val_dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 0.01
+  betas:
+  - 0.95
+  - 0.999
+  eps: 1.0e-08
+  weight_decay: 1.0e-06
+training:
+  device: cuda:0
+  seed: 45
+  debug: false
+  resume: false
+  lr_scheduler: cosine
+  lr_warmup_steps: 10
+  num_epochs: 20001
+  gradient_accumulate_every: 8
+  grad_norm_clip: 0.5
+  rollout_every: 100
+  checkpoint_every: 100
+  val_every: 1
+  sample_every: 100
+  sample_max_batch: 128
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-split-ablation
+  resume: true
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags:
+  - ${name}
+  - ${task_name}
+  - ${exp_name}
+  id: null
+  group: null
+checkpoint:
+  topk:
+    monitor_key: test_success_rate
+    mode: max
+    k: 5
+    format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+  save_last_ckpt: true
+  save_last_snapshot: false
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+task:
+  name: box-close-v2
+  obs_dim: 9
+  action_dim: 4
+  env_runner:
+    _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+    env_name: llf-metaworld-box-close-v2
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 30
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.9
+  dataset:
+    _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+    data_path: datasets/box-close-v2-split0.05.pt
+    data_path2: datasets/box-close-v2-split0.05.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: true
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: pretrained
+  finetune_mode: orig
+  checkpoint: data/outputs/2025.12.08/15.00.22_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+  max_length: 100
+  lora_config:
+    r: 32
+    lora_alpha: 64
+    lora_dropout: 0.05
+    bias: none
+    task_type: CAUSAL_LM
+  prompter:
+    _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}