Upload folder using huggingface_hub

.gitattributes

@@ -93,3 +93,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 2026.02.19/17.30.05_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 2026.02.19/17.30.05_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-8835/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 2026.02.19/17.30.05_train_llm_lowdim_box-close-v2/wandb/run-20260219_173015-8idvdxkw/run-8idvdxkw.wandb filter=lfs diff=lfs merge=lfs -text
+2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/run-a11bxnob.wandb filter=lfs diff=lfs merge=lfs -text
+2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/run-lro35ktw.wandb filter=lfs diff=lfs merge=lfs -text

2026.02.22/19.27.05_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: 42
+  debug: false
+  resume: false
+  lr_scheduler: cosine
+  lr_warmup_steps: 10
+  num_epochs: 1001
+  gradient_accumulate_every: 8
+  grad_norm_clip: 0.5
+  rollout_every: 5
+  checkpoint_every: 5
+  val_every: 1
+  sample_every: 5
+  sample_max_batch: 128
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-llm-backbone
+  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-o4-mini.pt
+    data_path2: datasets/box-close-v2-o4-mini.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: Qwen/Qwen2-0.5B-Instruct
+  model_name: Qwen2-0.5B-Instruct
+  config_target: llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config
+  causal_lm_target: llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: ete-finetuned
+  finetune_mode: orig
+  checkpoint: data/outputs/2026.02.20/17.05.08_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-17670
+  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.qwen2_prompter.Qwen2Prompter
+    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}

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/.hydra/hydra.yaml

@@ -0,0 +1,155 @@
+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:
+    - policy.loss_llm_weight=1.0e-3
+  job:
+    name: train
+    chdir: null
+    override_dirname: policy.loss_llm_weight=1.0e-3
+    id: ???
+    num: ???
+    config_name: llmbc_box-close-v2_o4-mini_qwen2-0.5b-instruct.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /work/u1131674/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /work/u1131674/LLM-BC/config/llm_backbone
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_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

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- policy.loss_llm_weight=1.0e-3

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0720-test_success_rate=0.920.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88b976264c9b37b924af4e72991c88ce2cb1f2c30504dd5d650fdb4dc4340cdc
+size 864520

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0730-test_success_rate=0.960.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f79151624af5e7ccbc1ecbfec40ec3497e5a0648514b4b3a4460a7257a3db0b
+size 864520

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0740-test_success_rate=0.920.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5cbfda0f52e7fba81575c166b48098ac09d94f6f3744c706f1534eb4b54d9bfe
+size 864520

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0755-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c076f335b702a8161b9b2960cdc0c88fe366fa6eb62dccbcc4c760cdf5fb3d9a
+size 864520

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0800-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:858485892db368ce72e18c5fbafaaae79d2a8fd3ccc3f91cb8ba5bb353e38f78
+size 864520

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f42e86a45d82cfecd618d149fd7e99d91054cc40ecf36e24914501176725e6f
+size 864520

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7401a21a6c51f5629984fc2009fdbb543a31d4560205bc6766db15f42441d4e9
+size 51875243

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/train.log

@@ -0,0 +1,8 @@
+[2026-02-22 19:27:07,429][numexpr.utils][INFO] - Note: detected 112 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable.
+[2026-02-22 19:27:07,429][numexpr.utils][INFO] - Note: NumExpr detected 112 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2026-02-22 19:27:07,429][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2026-02-22 19:27:14,029][datasets][INFO] - PyTorch version 2.2.2 available.
+[2026-02-22 19:27:14,030][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2026-02-22 19:27:14,031][datasets][INFO] - JAX version 0.4.30 available.
+[2026-02-22 19:30:20,255][absl][INFO] - MUJOCO_GL=osmesa, attempting to import specified OpenGL backend.
+[2026-02-22 19:30:20,790][absl][INFO] - MuJoCo library version is: 2.3.7

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2026-02-22T19:30:31.541837404+08:00","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2026-02-22T19:30:31.54184792+08:00","level":"INFO","msg":"created symlink","path":"/work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/logs/debug-core.log"}
+{"time":"2026-02-22T19:30:31.659377035+08:00","level":"INFO","msg":"created new stream","id":"a11bxnob"}
+{"time":"2026-02-22T19:30:31.659409289+08:00","level":"INFO","msg":"stream: started","id":"a11bxnob"}
+{"time":"2026-02-22T19:30:31.65943774+08:00","level":"INFO","msg":"sender: started","stream_id":"a11bxnob"}
+{"time":"2026-02-22T19:30:31.659425196+08:00","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"a11bxnob"}}
+{"time":"2026-02-22T19:30:31.659438281+08:00","level":"INFO","msg":"handler: started","stream_id":{"value":"a11bxnob"}}
+{"time":"2026-02-22T19:30:32.591004129+08:00","level":"INFO","msg":"Starting system monitor"}
+{"time":"2026-02-23T14:35:08.182480973+08:00","level":"INFO","msg":"stream: closing","id":"a11bxnob"}
+{"time":"2026-02-23T14:35:08.182527674+08:00","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2026-02-23T14:35:08.183145165+08:00","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2026-02-23T14:35:11.8322081+08:00","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2026-02-23T14:35:12.140214863+08:00","level":"INFO","msg":"handler: closed","stream_id":{"value":"a11bxnob"}}
+{"time":"2026-02-23T14:35:12.140243473+08:00","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"a11bxnob"}}
+{"time":"2026-02-23T14:35:12.140275962+08:00","level":"INFO","msg":"sender: closed","stream_id":"a11bxnob"}
+{"time":"2026-02-23T14:35:12.140309694+08:00","level":"INFO","msg":"stream: closed","id":"a11bxnob"}

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/debug.log

@@ -0,0 +1,27 @@
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Configure stats pid to 1326874
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Loading settings from /home/u1131674/.config/wandb/settings
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Loading settings from /work/u1131674/LLM-BC/wandb/settings
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/work/u1131674/LLM-BC/train.py', 'program': '/work/u1131674/LLM-BC/./train.py'}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Applying login settings: {}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:_log_setup():533] Logging user logs to /work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/logs/debug.log
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:_log_setup():534] Logging internal logs to /work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/logs/debug-internal.log
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:init():619] calling init triggers
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [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': 'Qwen/Qwen2-0.5B-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': 'Qwen/Qwen2-0.5B-Instruct', 'model_name': 'Qwen2-0.5B-Instruct', 'config_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config', 'causal_lm_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'ete-finetuned', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2026.02.20/17.05.08_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-17670', '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.qwen2_prompter.Qwen2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'Qwen/Qwen2-0.5B-Instruct'}, 'run': {'dir': 'data/outputs/2026.02.22/19.27.05_Qwen/Qwen2-0.5B-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': 42, 'debug': False, 'resume': False, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 10, 'num_epochs': 1001, 'gradient_accumulate_every': 8, 'grad_norm_clip': 0.5, 'rollout_every': 5, 'checkpoint_every': 5, 'val_every': 1, 'sample_every': 5, 'sample_max_batch': 128, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-llm-backbone', 'resume': True, 'mode': 'online', 'name': '2026.02.22-19.27.05_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/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2', 'wandb_name_base': '2026.02.22-19.27.05_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-o4-mini.pt', 'data_path2': 'datasets/box-close-v2-o4-mini.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': 'Qwen/Qwen2-0.5B-Instruct', 'model_name': 'Qwen2-0.5B-Instruct', 'config_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config', 'causal_lm_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'ete-finetuned', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2026.02.20/17.05.08_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-17670', '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.qwen2_prompter.Qwen2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'Qwen/Qwen2-0.5B-Instruct'}, 'run': {'dir': 'data/outputs/2026.02.22/19.27.05_Qwen/Qwen2-0.5B-Instruct'}}}}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:init():669] starting backend
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:init():673] sending inform_init request
+2026-02-22 19:30:31,539 INFO    MainThread:1326874 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2026-02-22 19:30:31,539 INFO    MainThread:1326874 [wandb_init.py:init():686] backend started and connected
+2026-02-22 19:30:31,597 INFO    MainThread:1326874 [wandb_init.py:init():781] updated telemetry
+2026-02-22 19:30:31,660 INFO    MainThread:1326874 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2026-02-22 19:30:32,585 INFO    MainThread:1326874 [wandb_init.py:init():867] starting run threads in backend
+2026-02-22 19:30:33,135 INFO    MainThread:1326874 [wandb_run.py:_console_start():2451] atexit reg
+2026-02-22 19:30:33,135 INFO    MainThread:1326874 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2026-02-22 19:30:33,135 INFO    MainThread:1326874 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2026-02-22 19:30:33,136 INFO    MainThread:1326874 [wandb_run.py:_redirect():2389] Redirects installed.
+2026-02-22 19:30:33,138 INFO    MainThread:1326874 [wandb_init.py:init():911] run started, returning control to user process
+2026-02-22 19:30:33,139 INFO    MainThread:1326874 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2'}
+2026-02-23 14:35:08,182 WARNING MsgRouterThr:1326874 [router.py:message_loop():75] message_loop has been closed

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/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.qwen2_lowdim_model.LowdimQwen2ForCausalLM
+        checkpoint: data/outputs/2026.02.20/17.05.08_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-17670
+        config_target: llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: Qwen/Qwen2-0.5B-Instruct
+            run:
+                dir: data/outputs/2026.02.22/19.27.05_Qwen/Qwen2-0.5B-Instruct
+        llm_mode: ete-finetuned
+        lora_config:
+            bias: none
+            lora_alpha: 64
+            lora_dropout: 0.05
+            r: 32
+            task_type: CAUSAL_LM
+        max_length: 100
+        model_name: Qwen2-0.5B-Instruct
+        name: Qwen/Qwen2-0.5B-Instruct
+        prompter:
+            _target_: llmbc.translator.prompter.qwen2_prompter.Qwen2Prompter
+            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: 2026.02.22-19.27.05_train_llmbc_lowdim_box-close-v2
+        project: box-close-v2-training-llm-backbone
+        resume: true
+        tags:
+            - train_llmbc_lowdim
+            - box-close-v2
+            - default
+model_name:
+    value: Qwen/Qwen2-0.5B-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2
+        wandb_name_base: 2026.02.22-19.27.05_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: /work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_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.qwen2_lowdim_model.LowdimQwen2ForCausalLM
+                    checkpoint: data/outputs/2026.02.20/17.05.08_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-17670
+                    config_target: llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: Qwen/Qwen2-0.5B-Instruct
+                        run:
+                            dir: data/outputs/2026.02.22/19.27.05_Qwen/Qwen2-0.5B-Instruct
+                    llm_mode: ete-finetuned
+                    lora_config:
+                        bias: none
+                        lora_alpha: 64
+                        lora_dropout: 0.05
+                        r: 32
+                        task_type: CAUSAL_LM
+                    max_length: 100
+                    model_name: Qwen2-0.5B-Instruct
+                    name: Qwen/Qwen2-0.5B-Instruct
+                    prompter:
+                        _target_: llmbc.translator.prompter.qwen2_prompter.Qwen2Prompter
+                        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-o4-mini.pt
+            data_path2: datasets/box-close-v2-o4-mini.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: 5
+        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: 1001
+        resume: false
+        rollout_every: 5
+        sample_every: 5
+        sample_max_batch: 128
+        seed: 42
+        tqdm_interval_sec: 1
+        val_every: 1
+val_dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: true

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/files/output.log

@@ -0,0 +1,3 @@
+Eval MetaworldLowdimRunner 1/6:   0%|          | 0/30 [00:00<?, ?it/s]/work/u1131674/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)
+                                                                                                        

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/files/wandb-metadata.json

@@ -0,0 +1,107 @@
+{
+  "os": "Linux-4.18.0-513.24.1.el8_9.x86_64-x86_64-with-glibc2.28",
+  "python": "3.9.25",
+  "startedAt": "2026-02-22T11:30:31.539797Z",
+  "args": [
+    "--config-path",
+    "config/llm_backbone",
+    "--config-name",
+    "llmbc_box-close-v2_o4-mini_qwen2-0.5b-instruct.yaml",
+    "policy.loss_llm_weight=1.0e-3"
+  ],
+  "program": "/work/u1131674/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "729cd85aa68ca8485e438979582080a467a43a2e"
+  },
+  "email": "chris920325@gmail.com",
+  "root": "/work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2",
+  "host": "hgpn02",
+  "username": "u1131674",
+  "executable": "/home/u1131674/.conda/envs/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 112,
+  "cpu_count_logical": 112,
+  "gpu": "NVIDIA H100 80GB HBM3",
+  "gpu_count": 1,
+  "disk": {
+    "/": {
+      "total": "1918024196096",
+      "used": "229413556224"
+    }
+  },
+  "memory": {
+    "total": "2163622887424"
+  },
+  "cpu": {
+    "count": 112,
+    "countLogical": 112
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    }
+  ],
+  "slurm": {
+    "cluster_name": "hpc",
+    "conf": "/etc/slurm/slurm.conf",
+    "cpu_bind": "quiet,mask_cpu:0x0000000FE0001000000000000000",
+    "cpu_bind_list": "0x0000000FE0001000000000000000",
+    "cpu_bind_type": "mask_cpu:",
+    "cpu_bind_verbose": "quiet",
+    "cpus_on_node": "8",
+    "cpus_per_task": "8",
+    "distribution": "cyclic,pack",
+    "gpus_on_node": "1",
+    "gpus_per_node": "1",
+    "gtids": "0",
+    "job_account": "mst114558",
+    "job_cpus_per_node": "8",
+    "job_end_time": "1771932422",
+    "job_gid": "106773",
+    "job_group": "MST114558",
+    "job_id": "122446",
+    "job_name": "python3",
+    "job_nodelist": "hgpn02",
+    "job_num_nodes": "1",
+    "job_partition": "normal",
+    "job_qos": "normal",
+    "job_start_time": "1771759622",
+    "job_uid": "41408",
+    "job_user": "u1131674",
+    "jobid": "122446",
+    "launch_node_ipaddr": "172.21.101.1",
+    "localid": "0",
+    "mem_per_node": "204800",
+    "nnodes": "1",
+    "nodeid": "0",
+    "nodelist": "hgpn02",
+    "nprocs": "1",
+    "ntasks": "1",
+    "prio_process": "0",
+    "procid": "0",
+    "srun_comm_host": "172.21.101.1",
+    "srun_comm_port": "46217",
+    "step_gpus": "5",
+    "step_id": "0",
+    "step_launcher_port": "46217",
+    "step_nodelist": "hgpn02",
+    "step_num_nodes": "1",
+    "step_num_tasks": "1",
+    "step_tasks_per_node": "1",
+    "stepid": "0",
+    "submit_dir": "/work/u1131674/LLM-BC",
+    "submit_host": "cbi-lgn01",
+    "task_pid": "1326874",
+    "tasks_per_node": "1",
+    "topology_addr": "ibsw1.hgpn02",
+    "topology_addr_pattern": "switch.node",
+    "tres_per_task": "cpu:8",
+    "umask": "0022"
+  },
+  "cudaVersion": "12.4"
+}

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/files/wandb-summary.json

@@ -0,0 +1 @@
+{"train_action_mse_error":6.035442856955342e-05,"test/success_rate":0.86,"_timestamp":1.7718285057607238e+09,"train_loss_bc":1.3953307643532753e-05,"test/mean_score":7.723069319548529,"global_step":238237,"train/cumulative_reward":16.62698265816893,"_runtime":68676.642713132,"_wandb":{"runtime":68676},"lr":2.7842694261259738e-11,"val_loss":0.0001193940406665206,"_step":238237,"epoch":1000,"test/cumulative_reward":14.384284103834462,"train_loss":0.00019799631605133263,"train/success_rate":1,"grad_norm":0.015387633815407753,"train/mean_score":8.674362803321433,"train_loss_llm":0}

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2026-02-22T19:30:30.587724868+08:00","level":"INFO","msg":"started logging, with flags","port-filename":"/tmp/tmpegnmnt6p/port-1326874.txt","pid":1326874,"debug":false,"disable-analytics":false}
+{"time":"2026-02-22T19:30:30.587750061+08:00","level":"INFO","msg":"FeatureState","shutdownOnParentExitEnabled":false}
+{"time":"2026-02-22T19:30:30.588303649+08:00","level":"INFO","msg":"Will exit if parent process dies.","ppid":1326874}
+{"time":"2026-02-22T19:30:30.588293422+08:00","level":"INFO","msg":"server is running","addr":{"IP":"127.0.0.1","Port":33019,"Zone":""}}
+{"time":"2026-02-22T19:30:30.780642064+08:00","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"127.0.0.1:38162"}
+{"time":"2026-02-22T19:30:31.540679098+08:00","level":"INFO","msg":"handleInformInit: received","streamId":"a11bxnob","id":"127.0.0.1:38162"}
+{"time":"2026-02-22T19:30:31.659414095+08:00","level":"INFO","msg":"handleInformInit: stream started","streamId":"a11bxnob","id":"127.0.0.1:38162"}
+{"time":"2026-02-23T14:35:08.182284859+08:00","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"127.0.0.1:38162"}
+{"time":"2026-02-23T14:35:08.182495978+08:00","level":"INFO","msg":"server is shutting down"}
+{"time":"2026-02-23T14:35:08.18248388+08:00","level":"INFO","msg":"connection: Close: initiating connection closure","id":"127.0.0.1:38162"}
+{"time":"2026-02-23T14:35:08.182581699+08:00","level":"INFO","msg":"connection: Close: connection successfully closed","id":"127.0.0.1:38162"}
+{"time":"2026-02-23T14:35:12.140585035+08:00","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"127.0.0.1:38162"}
+{"time":"2026-02-23T14:35:12.140598261+08:00","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"127.0.0.1:38162"}
+{"time":"2026-02-23T14:35:12.140606188+08:00","level":"INFO","msg":"server is closed"}

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/logs/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2026-02-22T19:30:31.541837404+08:00","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2026-02-22T19:30:31.54184792+08:00","level":"INFO","msg":"created symlink","path":"/work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/logs/debug-core.log"}
+{"time":"2026-02-22T19:30:31.659377035+08:00","level":"INFO","msg":"created new stream","id":"a11bxnob"}
+{"time":"2026-02-22T19:30:31.659409289+08:00","level":"INFO","msg":"stream: started","id":"a11bxnob"}
+{"time":"2026-02-22T19:30:31.65943774+08:00","level":"INFO","msg":"sender: started","stream_id":"a11bxnob"}
+{"time":"2026-02-22T19:30:31.659425196+08:00","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"a11bxnob"}}
+{"time":"2026-02-22T19:30:31.659438281+08:00","level":"INFO","msg":"handler: started","stream_id":{"value":"a11bxnob"}}
+{"time":"2026-02-22T19:30:32.591004129+08:00","level":"INFO","msg":"Starting system monitor"}
+{"time":"2026-02-23T14:35:08.182480973+08:00","level":"INFO","msg":"stream: closing","id":"a11bxnob"}
+{"time":"2026-02-23T14:35:08.182527674+08:00","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2026-02-23T14:35:08.183145165+08:00","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2026-02-23T14:35:11.8322081+08:00","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2026-02-23T14:35:12.140214863+08:00","level":"INFO","msg":"handler: closed","stream_id":{"value":"a11bxnob"}}
+{"time":"2026-02-23T14:35:12.140243473+08:00","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"a11bxnob"}}
+{"time":"2026-02-23T14:35:12.140275962+08:00","level":"INFO","msg":"sender: closed","stream_id":"a11bxnob"}
+{"time":"2026-02-23T14:35:12.140309694+08:00","level":"INFO","msg":"stream: closed","id":"a11bxnob"}

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/logs/debug.log

@@ -0,0 +1,27 @@
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Configure stats pid to 1326874
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Loading settings from /home/u1131674/.config/wandb/settings
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Loading settings from /work/u1131674/LLM-BC/wandb/settings
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/work/u1131674/LLM-BC/train.py', 'program': '/work/u1131674/LLM-BC/./train.py'}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_setup.py:_flush():79] Applying login settings: {}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:_log_setup():533] Logging user logs to /work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/logs/debug.log
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:_log_setup():534] Logging internal logs to /work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/logs/debug-internal.log
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:init():619] calling init triggers
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [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': 'Qwen/Qwen2-0.5B-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': 'Qwen/Qwen2-0.5B-Instruct', 'model_name': 'Qwen2-0.5B-Instruct', 'config_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config', 'causal_lm_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'ete-finetuned', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2026.02.20/17.05.08_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-17670', '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.qwen2_prompter.Qwen2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'Qwen/Qwen2-0.5B-Instruct'}, 'run': {'dir': 'data/outputs/2026.02.22/19.27.05_Qwen/Qwen2-0.5B-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': 42, 'debug': False, 'resume': False, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 10, 'num_epochs': 1001, 'gradient_accumulate_every': 8, 'grad_norm_clip': 0.5, 'rollout_every': 5, 'checkpoint_every': 5, 'val_every': 1, 'sample_every': 5, 'sample_max_batch': 128, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-llm-backbone', 'resume': True, 'mode': 'online', 'name': '2026.02.22-19.27.05_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/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2', 'wandb_name_base': '2026.02.22-19.27.05_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-o4-mini.pt', 'data_path2': 'datasets/box-close-v2-o4-mini.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': 'Qwen/Qwen2-0.5B-Instruct', 'model_name': 'Qwen2-0.5B-Instruct', 'config_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config', 'causal_lm_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'ete-finetuned', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2026.02.20/17.05.08_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-17670', '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.qwen2_prompter.Qwen2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'Qwen/Qwen2-0.5B-Instruct'}, 'run': {'dir': 'data/outputs/2026.02.22/19.27.05_Qwen/Qwen2-0.5B-Instruct'}}}}
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:init():669] starting backend
+2026-02-22 19:30:31,537 INFO    MainThread:1326874 [wandb_init.py:init():673] sending inform_init request
+2026-02-22 19:30:31,539 INFO    MainThread:1326874 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2026-02-22 19:30:31,539 INFO    MainThread:1326874 [wandb_init.py:init():686] backend started and connected
+2026-02-22 19:30:31,597 INFO    MainThread:1326874 [wandb_init.py:init():781] updated telemetry
+2026-02-22 19:30:31,660 INFO    MainThread:1326874 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2026-02-22 19:30:32,585 INFO    MainThread:1326874 [wandb_init.py:init():867] starting run threads in backend
+2026-02-22 19:30:33,135 INFO    MainThread:1326874 [wandb_run.py:_console_start():2451] atexit reg
+2026-02-22 19:30:33,135 INFO    MainThread:1326874 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2026-02-22 19:30:33,135 INFO    MainThread:1326874 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2026-02-22 19:30:33,136 INFO    MainThread:1326874 [wandb_run.py:_redirect():2389] Redirects installed.
+2026-02-22 19:30:33,138 INFO    MainThread:1326874 [wandb_init.py:init():911] run started, returning control to user process
+2026-02-22 19:30:33,139 INFO    MainThread:1326874 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/work/u1131674/LLM-BC/data/outputs/2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2'}
+2026-02-23 14:35:08,182 WARNING MsgRouterThr:1326874 [router.py:message_loop():75] message_loop has been closed

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_193031-a11bxnob/run-a11bxnob.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fe0705d48d9668156898404bfd36eca3ce340caecfa3d30e2e8cfa91c3145f9
+size 217021191

2026.02.22/19.27.05_train_llmbc_lowdim_box-close-v2/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "a11bxnob"}

2026.02.22/20.15.02_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: 42
+  debug: false
+  resume: false
+  lr_scheduler: cosine
+  lr_warmup_steps: 10
+  num_epochs: 1001
+  gradient_accumulate_every: 8
+  grad_norm_clip: 0.5
+  rollout_every: 5
+  checkpoint_every: 5
+  val_every: 1
+  sample_every: 5
+  sample_max_batch: 128
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-llm-backbone
+  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-o4-mini.pt
+    data_path2: datasets/box-close-v2-o4-mini.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: Qwen/Qwen2-0.5B-Instruct
+  model_name: Qwen2-0.5B-Instruct
+  config_target: llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config
+  causal_lm_target: llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: ete-finetuned
+  finetune_mode: orig
+  checkpoint: data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890
+  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.qwen2_prompter.Qwen2Prompter
+    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}

2026.02.22/20.15.02_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:
+    - policy.loss_llm_weight=1.0e-3
+    - llm.checkpoint=data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890
+  job:
+    name: train
+    chdir: null
+    override_dirname: llm.checkpoint=data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890,policy.loss_llm_weight=1.0e-3
+    id: ???
+    num: ???
+    config_name: llmbc_box-close-v2_o4-mini_qwen2-0.5b-instruct.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /work/u1131674/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /work/u1131674/LLM-BC/config/llm_backbone
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_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

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/.hydra/overrides.yaml

@@ -0,0 +1,2 @@
+- policy.loss_llm_weight=1.0e-3
+- llm.checkpoint=data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0395-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:632eb9a15b09e5f5c8a63c6db63acf11daafecc5b50b2a6805c0c5a02f9a52bc
+size 864520

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0540-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72aec2db8fe6f6f361af0cedaf0beb26e638e669bfd79d2589188e0bd60932bc
+size 864520

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0550-test_success_rate=0.900.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f466ed00c3adc635322ec202dc2e2613ded5e6b29ba998f615f7f2af8c770e37
+size 864520

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0755-test_success_rate=0.900.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69a6d983cffb0e6acdabd2944337933ce5846d4ac1652cd4855928c5e72010c3
+size 864520

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0895-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d94b218010275aa91ae13009c2c2950d69a6249e1b8a57757d2b60712473491e
+size 864520

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20484d14ee5bd26afcdd176f91145d263dc7209e604ce68266d81664ea22ad53
+size 864520

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41dd85e20be2ecf1ad726c2205728c9fc5551a06113f24aa3809c9522742a835
+size 50069975

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/train.log

@@ -0,0 +1,8 @@
+[2026-02-22 20:15:04,992][numexpr.utils][INFO] - Note: detected 112 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable.
+[2026-02-22 20:15:04,993][numexpr.utils][INFO] - Note: NumExpr detected 112 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2026-02-22 20:15:04,993][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2026-02-22 20:15:11,689][datasets][INFO] - PyTorch version 2.2.2 available.
+[2026-02-22 20:15:11,689][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2026-02-22 20:15:11,690][datasets][INFO] - JAX version 0.4.30 available.
+[2026-02-22 20:15:21,797][absl][INFO] - MUJOCO_GL=osmesa, attempting to import specified OpenGL backend.
+[2026-02-22 20:15:21,807][absl][INFO] - MuJoCo library version is: 2.3.7

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2026-02-22T20:15:22.796022055+08:00","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2026-02-22T20:15:22.796032572+08:00","level":"INFO","msg":"created symlink","path":"/work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/logs/debug-core.log"}
+{"time":"2026-02-22T20:15:22.913479056+08:00","level":"INFO","msg":"created new stream","id":"lro35ktw"}
+{"time":"2026-02-22T20:15:22.913512261+08:00","level":"INFO","msg":"stream: started","id":"lro35ktw"}
+{"time":"2026-02-22T20:15:22.913552888+08:00","level":"INFO","msg":"sender: started","stream_id":"lro35ktw"}
+{"time":"2026-02-22T20:15:22.91353557+08:00","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"lro35ktw"}}
+{"time":"2026-02-22T20:15:22.913539052+08:00","level":"INFO","msg":"handler: started","stream_id":{"value":"lro35ktw"}}
+{"time":"2026-02-22T20:15:23.68138796+08:00","level":"INFO","msg":"Starting system monitor"}
+{"time":"2026-02-23T15:27:44.252194309+08:00","level":"INFO","msg":"stream: closing","id":"lro35ktw"}
+{"time":"2026-02-23T15:27:44.25223112+08:00","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2026-02-23T15:27:44.252656758+08:00","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2026-02-23T15:27:45.432756272+08:00","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2026-02-23T15:27:45.711101559+08:00","level":"INFO","msg":"handler: closed","stream_id":{"value":"lro35ktw"}}
+{"time":"2026-02-23T15:27:45.711139656+08:00","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"lro35ktw"}}
+{"time":"2026-02-23T15:27:45.711162708+08:00","level":"INFO","msg":"sender: closed","stream_id":"lro35ktw"}
+{"time":"2026-02-23T15:27:45.711193754+08:00","level":"INFO","msg":"stream: closed","id":"lro35ktw"}

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/debug.log

@@ -0,0 +1,27 @@
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Configure stats pid to 3939074
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Loading settings from /home/u1131674/.config/wandb/settings
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Loading settings from /work/u1131674/LLM-BC/wandb/settings
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/work/u1131674/LLM-BC/train.py', 'program': '/work/u1131674/LLM-BC/./train.py'}
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Applying login settings: {}
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:_log_setup():533] Logging user logs to /work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/logs/debug.log
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:_log_setup():534] Logging internal logs to /work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/logs/debug-internal.log
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:init():619] calling init triggers
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [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': 'Qwen/Qwen2-0.5B-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': 'Qwen/Qwen2-0.5B-Instruct', 'model_name': 'Qwen2-0.5B-Instruct', 'config_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config', 'causal_lm_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'ete-finetuned', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890', '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.qwen2_prompter.Qwen2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'Qwen/Qwen2-0.5B-Instruct'}, 'run': {'dir': 'data/outputs/2026.02.22/20.15.02_Qwen/Qwen2-0.5B-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': 42, 'debug': False, 'resume': False, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 10, 'num_epochs': 1001, 'gradient_accumulate_every': 8, 'grad_norm_clip': 0.5, 'rollout_every': 5, 'checkpoint_every': 5, 'val_every': 1, 'sample_every': 5, 'sample_max_batch': 128, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-llm-backbone', 'resume': True, 'mode': 'online', 'name': '2026.02.22-20.15.02_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/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2', 'wandb_name_base': '2026.02.22-20.15.02_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-o4-mini.pt', 'data_path2': 'datasets/box-close-v2-o4-mini.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': 'Qwen/Qwen2-0.5B-Instruct', 'model_name': 'Qwen2-0.5B-Instruct', 'config_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config', 'causal_lm_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'ete-finetuned', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890', '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.qwen2_prompter.Qwen2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'Qwen/Qwen2-0.5B-Instruct'}, 'run': {'dir': 'data/outputs/2026.02.22/20.15.02_Qwen/Qwen2-0.5B-Instruct'}}}}
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:init():669] starting backend
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:init():673] sending inform_init request
+2026-02-22 20:15:22,793 INFO    MainThread:3939074 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2026-02-22 20:15:22,794 INFO    MainThread:3939074 [wandb_init.py:init():686] backend started and connected
+2026-02-22 20:15:22,803 INFO    MainThread:3939074 [wandb_init.py:init():781] updated telemetry
+2026-02-22 20:15:22,862 INFO    MainThread:3939074 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2026-02-22 20:15:23,677 INFO    MainThread:3939074 [wandb_init.py:init():867] starting run threads in backend
+2026-02-22 20:15:24,224 INFO    MainThread:3939074 [wandb_run.py:_console_start():2451] atexit reg
+2026-02-22 20:15:24,224 INFO    MainThread:3939074 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2026-02-22 20:15:24,224 INFO    MainThread:3939074 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2026-02-22 20:15:24,224 INFO    MainThread:3939074 [wandb_run.py:_redirect():2389] Redirects installed.
+2026-02-22 20:15:24,227 INFO    MainThread:3939074 [wandb_init.py:init():911] run started, returning control to user process
+2026-02-22 20:15:24,227 INFO    MainThread:3939074 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2'}
+2026-02-23 15:27:44,252 WARNING MsgRouterThr:3939074 [router.py:message_loop():75] message_loop has been closed

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/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.qwen2_lowdim_model.LowdimQwen2ForCausalLM
+        checkpoint: data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890
+        config_target: llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: Qwen/Qwen2-0.5B-Instruct
+            run:
+                dir: data/outputs/2026.02.22/20.15.02_Qwen/Qwen2-0.5B-Instruct
+        llm_mode: ete-finetuned
+        lora_config:
+            bias: none
+            lora_alpha: 64
+            lora_dropout: 0.05
+            r: 32
+            task_type: CAUSAL_LM
+        max_length: 100
+        model_name: Qwen2-0.5B-Instruct
+        name: Qwen/Qwen2-0.5B-Instruct
+        prompter:
+            _target_: llmbc.translator.prompter.qwen2_prompter.Qwen2Prompter
+            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: 2026.02.22-20.15.02_train_llmbc_lowdim_box-close-v2
+        project: box-close-v2-training-llm-backbone
+        resume: true
+        tags:
+            - train_llmbc_lowdim
+            - box-close-v2
+            - default
+model_name:
+    value: Qwen/Qwen2-0.5B-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2
+        wandb_name_base: 2026.02.22-20.15.02_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: /work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_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.qwen2_lowdim_model.LowdimQwen2ForCausalLM
+                    checkpoint: data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890
+                    config_target: llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: Qwen/Qwen2-0.5B-Instruct
+                        run:
+                            dir: data/outputs/2026.02.22/20.15.02_Qwen/Qwen2-0.5B-Instruct
+                    llm_mode: ete-finetuned
+                    lora_config:
+                        bias: none
+                        lora_alpha: 64
+                        lora_dropout: 0.05
+                        r: 32
+                        task_type: CAUSAL_LM
+                    max_length: 100
+                    model_name: Qwen2-0.5B-Instruct
+                    name: Qwen/Qwen2-0.5B-Instruct
+                    prompter:
+                        _target_: llmbc.translator.prompter.qwen2_prompter.Qwen2Prompter
+                        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-o4-mini.pt
+            data_path2: datasets/box-close-v2-o4-mini.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: 5
+        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: 1001
+        resume: false
+        rollout_every: 5
+        sample_every: 5
+        sample_max_batch: 128
+        seed: 42
+        tqdm_interval_sec: 1
+        val_every: 1
+val_dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: true

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/files/output.log

@@ -0,0 +1,3 @@
+Eval MetaworldLowdimRunner 1/6:   0%|          | 0/30 [00:00<?, ?it/s]/work/u1131674/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)
+                                                                                                        

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/files/wandb-metadata.json

@@ -0,0 +1,108 @@
+{
+  "os": "Linux-4.18.0-513.24.1.el8_9.x86_64-x86_64-with-glibc2.28",
+  "python": "3.9.25",
+  "startedAt": "2026-02-22T12:15:22.794314Z",
+  "args": [
+    "--config-path",
+    "config/llm_backbone",
+    "--config-name",
+    "llmbc_box-close-v2_o4-mini_qwen2-0.5b-instruct.yaml",
+    "policy.loss_llm_weight=1.0e-3",
+    "llm.checkpoint=data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890"
+  ],
+  "program": "/work/u1131674/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "729cd85aa68ca8485e438979582080a467a43a2e"
+  },
+  "email": "chris920325@gmail.com",
+  "root": "/work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2",
+  "host": "hgpn17",
+  "username": "u1131674",
+  "executable": "/home/u1131674/.conda/envs/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 112,
+  "cpu_count_logical": 112,
+  "gpu": "NVIDIA H100 80GB HBM3",
+  "gpu_count": 1,
+  "disk": {
+    "/": {
+      "total": "1918024196096",
+      "used": "199957532672"
+    }
+  },
+  "memory": {
+    "total": "2163677069312"
+  },
+  "cpu": {
+    "count": 112,
+    "countLogical": 112
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    }
+  ],
+  "slurm": {
+    "cluster_name": "hpc",
+    "conf": "/etc/slurm/slurm.conf",
+    "cpu_bind": "quiet,mask_cpu:0x0000E01E00008000000000000000",
+    "cpu_bind_list": "0x0000E01E00008000000000000000",
+    "cpu_bind_type": "mask_cpu:",
+    "cpu_bind_verbose": "quiet",
+    "cpus_on_node": "8",
+    "cpus_per_task": "8",
+    "distribution": "cyclic,pack",
+    "gpus_on_node": "1",
+    "gpus_per_node": "1",
+    "gtids": "0",
+    "job_account": "mst114558",
+    "job_cpus_per_node": "8",
+    "job_end_time": "1771935300",
+    "job_gid": "106773",
+    "job_group": "MST114558",
+    "job_id": "122457",
+    "job_name": "python3",
+    "job_nodelist": "hgpn17",
+    "job_num_nodes": "1",
+    "job_partition": "normal",
+    "job_qos": "normal",
+    "job_start_time": "1771762500",
+    "job_uid": "41408",
+    "job_user": "u1131674",
+    "jobid": "122457",
+    "launch_node_ipaddr": "172.21.101.1",
+    "localid": "0",
+    "mem_per_node": "204800",
+    "nnodes": "1",
+    "nodeid": "0",
+    "nodelist": "hgpn17",
+    "nprocs": "1",
+    "ntasks": "1",
+    "prio_process": "0",
+    "procid": "0",
+    "srun_comm_host": "172.21.101.1",
+    "srun_comm_port": "41731",
+    "step_gpus": "6",
+    "step_id": "0",
+    "step_launcher_port": "41731",
+    "step_nodelist": "hgpn17",
+    "step_num_nodes": "1",
+    "step_num_tasks": "1",
+    "step_tasks_per_node": "1",
+    "stepid": "0",
+    "submit_dir": "/work/u1131674/LLM-BC",
+    "submit_host": "cbi-lgn01",
+    "task_pid": "3939074",
+    "tasks_per_node": "1",
+    "topology_addr": "ibsw1.hgpn17",
+    "topology_addr_pattern": "switch.node",
+    "tres_per_task": "cpu:8",
+    "umask": "0022"
+  },
+  "cudaVersion": "12.4"
+}

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/files/wandb-summary.json

@@ -0,0 +1 @@
+{"train/mean_score":8.232452605194261,"grad_norm":0.010613621212542057,"_timestamp":1.771831661933487e+09,"train_loss":0.0001466139535530968,"test/cumulative_reward":12.538765441213028,"train/success_rate":0.9,"_step":238237,"_runtime":69141.457898941,"test/mean_score":6.13911056910327,"_wandb":{"runtime":69141},"epoch":1000,"train_loss_bc":1.851748675107956e-05,"train_action_mse_error":9.579761535860598e-05,"val_loss":6.219484203029424e-05,"test/success_rate":0.66,"train/cumulative_reward":16.15837920137297,"lr":2.7842694261259738e-11,"train_loss_llm":0,"global_step":238237}

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2026-02-22T20:15:22.192213059+08:00","level":"INFO","msg":"started logging, with flags","port-filename":"/tmp/tmpts3tst5d/port-3939074.txt","pid":3939074,"debug":false,"disable-analytics":false}
+{"time":"2026-02-22T20:15:22.19223562+08:00","level":"INFO","msg":"FeatureState","shutdownOnParentExitEnabled":false}
+{"time":"2026-02-22T20:15:22.192689502+08:00","level":"INFO","msg":"Will exit if parent process dies.","ppid":3939074}
+{"time":"2026-02-22T20:15:22.192683757+08:00","level":"INFO","msg":"server is running","addr":{"IP":"127.0.0.1","Port":35777,"Zone":""}}
+{"time":"2026-02-22T20:15:22.385461166+08:00","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"127.0.0.1:36806"}
+{"time":"2026-02-22T20:15:22.794952306+08:00","level":"INFO","msg":"handleInformInit: received","streamId":"lro35ktw","id":"127.0.0.1:36806"}
+{"time":"2026-02-22T20:15:22.913517455+08:00","level":"INFO","msg":"handleInformInit: stream started","streamId":"lro35ktw","id":"127.0.0.1:36806"}
+{"time":"2026-02-23T15:27:44.252134185+08:00","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"127.0.0.1:36806"}
+{"time":"2026-02-23T15:27:44.252198317+08:00","level":"INFO","msg":"server is shutting down"}
+{"time":"2026-02-23T15:27:44.252193387+08:00","level":"INFO","msg":"connection: Close: initiating connection closure","id":"127.0.0.1:36806"}
+{"time":"2026-02-23T15:27:44.252273838+08:00","level":"INFO","msg":"connection: Close: connection successfully closed","id":"127.0.0.1:36806"}
+{"time":"2026-02-23T15:27:45.711431536+08:00","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"127.0.0.1:36806"}
+{"time":"2026-02-23T15:27:45.711449468+08:00","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"127.0.0.1:36806"}
+{"time":"2026-02-23T15:27:45.71145982+08:00","level":"INFO","msg":"server is closed"}

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/logs/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2026-02-22T20:15:22.796022055+08:00","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2026-02-22T20:15:22.796032572+08:00","level":"INFO","msg":"created symlink","path":"/work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/logs/debug-core.log"}
+{"time":"2026-02-22T20:15:22.913479056+08:00","level":"INFO","msg":"created new stream","id":"lro35ktw"}
+{"time":"2026-02-22T20:15:22.913512261+08:00","level":"INFO","msg":"stream: started","id":"lro35ktw"}
+{"time":"2026-02-22T20:15:22.913552888+08:00","level":"INFO","msg":"sender: started","stream_id":"lro35ktw"}
+{"time":"2026-02-22T20:15:22.91353557+08:00","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"lro35ktw"}}
+{"time":"2026-02-22T20:15:22.913539052+08:00","level":"INFO","msg":"handler: started","stream_id":{"value":"lro35ktw"}}
+{"time":"2026-02-22T20:15:23.68138796+08:00","level":"INFO","msg":"Starting system monitor"}
+{"time":"2026-02-23T15:27:44.252194309+08:00","level":"INFO","msg":"stream: closing","id":"lro35ktw"}
+{"time":"2026-02-23T15:27:44.25223112+08:00","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2026-02-23T15:27:44.252656758+08:00","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2026-02-23T15:27:45.432756272+08:00","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2026-02-23T15:27:45.711101559+08:00","level":"INFO","msg":"handler: closed","stream_id":{"value":"lro35ktw"}}
+{"time":"2026-02-23T15:27:45.711139656+08:00","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"lro35ktw"}}
+{"time":"2026-02-23T15:27:45.711162708+08:00","level":"INFO","msg":"sender: closed","stream_id":"lro35ktw"}
+{"time":"2026-02-23T15:27:45.711193754+08:00","level":"INFO","msg":"stream: closed","id":"lro35ktw"}

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/logs/debug.log

@@ -0,0 +1,27 @@
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Configure stats pid to 3939074
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Loading settings from /home/u1131674/.config/wandb/settings
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Loading settings from /work/u1131674/LLM-BC/wandb/settings
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/work/u1131674/LLM-BC/train.py', 'program': '/work/u1131674/LLM-BC/./train.py'}
+2026-02-22 20:15:22,791 INFO    MainThread:3939074 [wandb_setup.py:_flush():79] Applying login settings: {}
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:_log_setup():533] Logging user logs to /work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/logs/debug.log
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:_log_setup():534] Logging internal logs to /work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/logs/debug-internal.log
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:init():619] calling init triggers
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [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': 'Qwen/Qwen2-0.5B-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': 'Qwen/Qwen2-0.5B-Instruct', 'model_name': 'Qwen2-0.5B-Instruct', 'config_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config', 'causal_lm_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'ete-finetuned', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890', '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.qwen2_prompter.Qwen2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'Qwen/Qwen2-0.5B-Instruct'}, 'run': {'dir': 'data/outputs/2026.02.22/20.15.02_Qwen/Qwen2-0.5B-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': 42, 'debug': False, 'resume': False, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 10, 'num_epochs': 1001, 'gradient_accumulate_every': 8, 'grad_norm_clip': 0.5, 'rollout_every': 5, 'checkpoint_every': 5, 'val_every': 1, 'sample_every': 5, 'sample_max_batch': 128, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-llm-backbone', 'resume': True, 'mode': 'online', 'name': '2026.02.22-20.15.02_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/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2', 'wandb_name_base': '2026.02.22-20.15.02_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-o4-mini.pt', 'data_path2': 'datasets/box-close-v2-o4-mini.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': 'Qwen/Qwen2-0.5B-Instruct', 'model_name': 'Qwen2-0.5B-Instruct', 'config_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2Config', 'causal_lm_target': 'llmbc.model.llm.qwen2_lowdim_model.LowdimQwen2ForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'ete-finetuned', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2026.02.16/02.01.01_train_llm_lowdim_box-close-v2/Qwen/Qwen2-0.5B-Instruct-finetuned-box-close-v2/checkpoint-5890', '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.qwen2_prompter.Qwen2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'Qwen/Qwen2-0.5B-Instruct'}, 'run': {'dir': 'data/outputs/2026.02.22/20.15.02_Qwen/Qwen2-0.5B-Instruct'}}}}
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:init():669] starting backend
+2026-02-22 20:15:22,792 INFO    MainThread:3939074 [wandb_init.py:init():673] sending inform_init request
+2026-02-22 20:15:22,793 INFO    MainThread:3939074 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2026-02-22 20:15:22,794 INFO    MainThread:3939074 [wandb_init.py:init():686] backend started and connected
+2026-02-22 20:15:22,803 INFO    MainThread:3939074 [wandb_init.py:init():781] updated telemetry
+2026-02-22 20:15:22,862 INFO    MainThread:3939074 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2026-02-22 20:15:23,677 INFO    MainThread:3939074 [wandb_init.py:init():867] starting run threads in backend
+2026-02-22 20:15:24,224 INFO    MainThread:3939074 [wandb_run.py:_console_start():2451] atexit reg
+2026-02-22 20:15:24,224 INFO    MainThread:3939074 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2026-02-22 20:15:24,224 INFO    MainThread:3939074 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2026-02-22 20:15:24,224 INFO    MainThread:3939074 [wandb_run.py:_redirect():2389] Redirects installed.
+2026-02-22 20:15:24,227 INFO    MainThread:3939074 [wandb_init.py:init():911] run started, returning control to user process
+2026-02-22 20:15:24,227 INFO    MainThread:3939074 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/work/u1131674/LLM-BC/data/outputs/2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2'}
+2026-02-23 15:27:44,252 WARNING MsgRouterThr:3939074 [router.py:message_loop():75] message_loop has been closed

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/run-20260222_201522-lro35ktw/run-lro35ktw.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6aea4e38be31950b196b2c9f600efe9a317c63185a85d8676c23258a5c3ae281
+size 213903416

2026.02.22/20.15.02_train_llmbc_lowdim_box-close-v2/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "lro35ktw"}