Upload folder using huggingface_hub

.gitattributes

@@ -419,3 +419,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 2026.05.04/15.27.22_train_llmbc_lowdim_box-close-v2/wandb/run-20260504_152808-z0p0su1a/run-z0p0su1a.wandb filter=lfs diff=lfs merge=lfs -text
 2026.05.04/17.10.49_train_llmbc_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
 2026.05.04/17.10.49_train_llmbc_lowdim_box-close-v2/wandb/run-20260504_171137-ewv8lvf2/run-ewv8lvf2.wandb filter=lfs diff=lfs merge=lfs -text
+2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_083754-304tn7ln/run-304tn7ln.wandb filter=lfs diff=lfs merge=lfs -text
+2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_084911-zvpenr2j/run-zvpenr2j.wandb filter=lfs diff=lfs merge=lfs -text

2026.05.05/08.37.08_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
+  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-human.pt
+    data_path2: datasets/box-close-v2-human.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: true
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: ete-finetuned
+  finetune_mode: orig
+  checkpoint: data/outputs/2026.05.04/12.16.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-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.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/.hydra/hydra.yaml

@@ -0,0 +1,156 @@
+hydra:
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - training.seed=42
+    - policy.loss_llm_weight=1.0e-3
+  job:
+    name: train
+    chdir: null
+    override_dirname: policy.loss_llm_weight=1.0e-3,training.seed=42
+    id: ???
+    num: ???
+    config_name: llmbc_box-close-v2_human.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /workspace/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /workspace/LLM-BC/config/lang_source
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/LLM-BC/data/outputs/2026.05.05/08.37.08_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

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/.hydra/overrides.yaml

@@ -0,0 +1,2 @@
+- training.seed=42
+- policy.loss_llm_weight=1.0e-3

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0165-test_success_rate=0.860.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b0c9b6ff114f562b663c673437a8e96a8a3bfb5c5c1fd51a4b387a2648bee13
+size 864520

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0365-test_success_rate=0.820.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:041555ac75be099073b655c33c5685baae8f143cfb45226ee4cee7404dcf2120
+size 864520

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0375-test_success_rate=0.840.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa78f7fc8d08bde7a2b3f05c0a3b23b56b2ef9f2bc3ee57b9ee0e818ba66c09a
+size 864520

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0410-test_success_rate=0.860.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1012bca590a626ad89ba0a99ce28103c24081091ba191b6eaa0c807139fa29be
+size 864520

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0480-test_success_rate=0.820.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ffe0910951b23f3bc220181f26abd7d968b73eb64f9eabddee5e22e6916bcf3
+size 864520

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e24deeed950b1cece412164f3ef43ef12799c37587f4aba8c875ed28a1538229
+size 864520

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:066dda143579460c7e9b10bc5de8faed0f2913a6b71d9df13d2b4f67b8831a7f
+size 51875524

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/train.log

@@ -0,0 +1,10 @@
+[2026-05-05 08:37:10,238][numexpr.utils][INFO] - Note: NumExpr detected 36 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2026-05-05 08:37:10,238][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2026-05-05 08:37:13,925][datasets][INFO] - PyTorch version 2.2.2 available.
+[2026-05-05 08:37:13,926][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2026-05-05 08:37:13,926][datasets][INFO] - JAX version 0.4.30 available.
+[2026-05-05 08:37:18,863][root][INFO] - running build_ext
+[2026-05-05 08:37:18,865][root][INFO] - building 'mujoco_py.cymj' extension
+[2026-05-05 08:37:18,866][root][INFO] - gcc -pthread -B /venv/llm-bc/compiler_compat -Wno-unused-result -Wsign-compare -DNDEBUG -O2 -Wall -fPIC -O2 -isystem /venv/llm-bc/include -I/venv/llm-bc/include -fPIC -O2 -isystem /venv/llm-bc/include -fPIC -I/venv/llm-bc/lib/python3.9/site-packages/mujoco_py -I/root/.mujoco/mujoco210/include -I/venv/llm-bc/lib/python3.9/site-packages/numpy/core/include -I/venv/llm-bc/include/python3.9 -c /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/cymj.c -o /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/temp.linux-x86_64-cpython-39/venv/llm-bc/lib/python3.9/site-packages/mujoco_py/cymj.o -fopenmp -w
+[2026-05-05 08:37:46,510][root][INFO] - gcc -pthread -B /venv/llm-bc/compiler_compat -Wno-unused-result -Wsign-compare -DNDEBUG -O2 -Wall -fPIC -O2 -isystem /venv/llm-bc/include -I/venv/llm-bc/include -fPIC -O2 -isystem /venv/llm-bc/include -fPIC -I/venv/llm-bc/lib/python3.9/site-packages/mujoco_py -I/root/.mujoco/mujoco210/include -I/venv/llm-bc/lib/python3.9/site-packages/numpy/core/include -I/venv/llm-bc/include/python3.9 -c /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/gl/osmesashim.c -o /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/temp.linux-x86_64-cpython-39/venv/llm-bc/lib/python3.9/site-packages/mujoco_py/gl/osmesashim.o -fopenmp -w
+[2026-05-05 08:37:46,621][root][INFO] - gcc -pthread -B /venv/llm-bc/compiler_compat -shared -Wl,-rpath,/venv/llm-bc/lib -Wl,-rpath-link,/venv/llm-bc/lib -L/venv/llm-bc/lib -L/venv/llm-bc/lib -Wl,-rpath,/venv/llm-bc/lib -Wl,-rpath-link,/venv/llm-bc/lib -L/venv/llm-bc/lib /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/temp.linux-x86_64-cpython-39/venv/llm-bc/lib/python3.9/site-packages/mujoco_py/cymj.o /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/temp.linux-x86_64-cpython-39/venv/llm-bc/lib/python3.9/site-packages/mujoco_py/gl/osmesashim.o -L/root/.mujoco/mujoco210/bin -Wl,--enable-new-dtags,-rpath,/root/.mujoco/mujoco210/bin -lmujoco210 -lglewosmesa -lOSMesa -lGL -o /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/lib.linux-x86_64-cpython-39/mujoco_py/cymj.cpython-39-x86_64-linux-gnu.so -fopenmp

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_083754-304tn7ln/files/config.yaml

@@ -0,0 +1,271 @@
+_target_:
+    value: llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace
+_wandb:
+    value:
+        cli_version: 0.18.6
+        m: []
+        python_version: 3.9.25
+        t:
+            "1":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 11
+                - 12
+                - 41
+                - 49
+                - 50
+                - 51
+                - 53
+                - 55
+                - 71
+                - 83
+                - 95
+                - 98
+                - 100
+                - 105
+            "2":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 11
+                - 12
+                - 41
+                - 49
+                - 50
+                - 51
+                - 53
+                - 55
+                - 71
+                - 83
+                - 95
+                - 98
+                - 100
+                - 105
+            "3":
+                - 13
+                - 15
+                - 16
+                - 23
+                - 55
+                - 61
+            "4": 3.9.25
+            "5": 0.18.6
+            "6": 4.47.1
+            "8":
+                - 5
+            "12": 0.18.6
+            "13": linux-x86_64
+action_dim:
+    value: 4
+checkpoint:
+    value:
+        save_last_ckpt: true
+        save_last_snapshot: false
+        topk:
+            format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+            k: 5
+            mode: max
+            monitor_key: test_success_rate
+dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: true
+exp_name:
+    value: default
+horizon:
+    value: 1
+llm:
+    value:
+        causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+        checkpoint: data/outputs/2026.05.04/12.16.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/checkpoint-5890
+        config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+            run:
+                dir: data/outputs/2026.05.05/08.37.08_HuggingFaceTB/SmolLM2-135M-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: SmolLM2-135M-Instruct
+        name: HuggingFaceTB/SmolLM2-135M-Instruct
+        prompter:
+            _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+            use_joint_mlp_projector: true
+        use_joint_mlp_projector: true
+        use_quantization: false
+llm_do_sample:
+    value: false
+llm_orig_expert_feedback:
+    value: true
+logging:
+    value:
+        group: null
+        id: null
+        mode: online
+        name: 2026.05.05-08.37.08_train_llmbc_lowdim_box-close-v2
+        project: box-close-v2-training
+        resume: true
+        tags:
+            - train_llmbc_lowdim
+            - box-close-v2
+            - default
+model_name:
+    value: HuggingFaceTB/SmolLM2-135M-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2
+        wandb_name_base: 2026.05.05-08.37.08_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: /workspace/LLM-BC/data/outputs/2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2
+past_action_visible:
+    value: false
+policy:
+    value:
+        _target_: llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy
+        action_dim: 4
+        horizon: 1
+        llm_discriminator:
+            _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+            llm_translator:
+                _target_: llmbc.translator.llm_translator.LLMTranslator
+                action_dim: 4
+                cfg:
+                    causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+                    checkpoint: data/outputs/2026.05.04/12.16.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/checkpoint-5890
+                    config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+                        run:
+                            dir: data/outputs/2026.05.05/08.37.08_HuggingFaceTB/SmolLM2-135M-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: SmolLM2-135M-Instruct
+                    name: HuggingFaceTB/SmolLM2-135M-Instruct
+                    prompter:
+                        _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+                        use_joint_mlp_projector: true
+                    use_joint_mlp_projector: true
+                    use_quantization: false
+                horizon: 1
+                n_action_steps: 1
+                n_obs_steps: 1
+                obs_dim: 9
+            task_id: box-close-v2
+        loss_bc_weight: 1
+        loss_llm_weight: 0.001
+        model:
+            _target_: llmbc.model.policy.policy_mlp.PolicyMLP
+            activation: relu
+            hidden_size:
+                - 256
+                - 256
+            input_size: 9
+            n_action_steps: 1
+            n_obs_steps: 1
+            output_size: 4
+        n_action_steps: 1
+        n_obs_steps: 1
+        normalize_llm_loss: true
+        obs_dim: 9
+task:
+    value:
+        action_dim: 4
+        dataset:
+            _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+            data_path: datasets/box-close-v2-human.pt
+            data_path2: datasets/box-close-v2-human.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

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_083754-304tn7ln/files/output.log

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

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_083754-304tn7ln/files/wandb-metadata.json

@@ -0,0 +1,56 @@
+{
+  "os": "Linux-5.15.0-140-generic-x86_64-with-glibc2.35",
+  "python": "3.9.25",
+  "startedAt": "2026-05-05T08:37:54.702009Z",
+  "args": [
+    "--config-path",
+    "config/lang_source",
+    "--config-name",
+    "llmbc_box-close-v2_human.yaml",
+    "training.seed=42",
+    "policy.loss_llm_weight=1.0e-3"
+  ],
+  "program": "/workspace/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "289618e617b0141d7a8a634ebe196cba52d07960"
+  },
+  "root": "/workspace/LLM-BC/data/outputs/2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2",
+  "host": "d8640274e028",
+  "username": "root",
+  "executable": "/venv/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 36,
+  "cpu_count_logical": 36,
+  "gpu": "NVIDIA GeForce RTX 4090",
+  "gpu_count": 2,
+  "disk": {
+    "/": {
+      "total": "569083166720",
+      "used": "425896898560"
+    }
+  },
+  "memory": {
+    "total": "135036993536"
+  },
+  "cpu": {
+    "count": 36,
+    "countLogical": 36
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA GeForce RTX 4090",
+      "memoryTotal": "25757220864",
+      "cudaCores": 16384,
+      "architecture": "Ada"
+    },
+    {
+      "name": "NVIDIA GeForce RTX 4090",
+      "memoryTotal": "25757220864",
+      "cudaCores": 16384,
+      "architecture": "Ada"
+    }
+  ],
+  "cudaVersion": "12.4"
+}

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_083754-304tn7ln/files/wandb-summary.json

@@ -0,0 +1 @@
+{"global_step":238237,"train/cumulative_reward":16.458843414359475,"_step":238237,"test/success_rate":0.68,"train_loss_bc":6.496794958366081e-05,"lr":2.7842694261259738e-11,"epoch":1000,"grad_norm":0.004614791367202997,"test/cumulative_reward":12.079566473088516,"test/mean_score":5.947032821378067,"train_action_mse_error":0.00011052155605284497,"_wandb":{"runtime":47759},"train_loss":0.00015682101172555916,"_runtime":47759.015233076,"val_loss":8.246159268310294e-05,"train/success_rate":1,"_timestamp":1.7780180336923614e+09,"train/mean_score":8.781898362910782,"train_loss_llm":0}

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_083754-304tn7ln/run-304tn7ln.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9393063fd24b6b291a5b4cee4d8cdb91d380d4e4a1eef8a5b9ad36a28fa498f2
+size 199879391

2026.05.05/08.37.08_train_llmbc_lowdim_box-close-v2/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "304tn7ln"}

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/.hydra/config.yaml

@@ -0,0 +1,163 @@
+name: train_llmbc_lowdim
+_target_: llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace
+obs_dim: ${task.obs_dim}
+action_dim: ${task.action_dim}
+task_name: ${task.name}
+exp_name: default
+model_name: ${llm.name}
+horizon: 1
+n_obs_steps: 1
+n_action_steps: 1
+n_latency_steps: 0
+past_action_visible: false
+llm_orig_expert_feedback: true
+llm_do_sample: false
+policy:
+  _target_: llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy
+  model:
+    _target_: llmbc.model.policy.policy_mlp.PolicyMLP
+    input_size: ${eval:'${n_obs_steps}*${obs_dim}'}
+    hidden_size:
+    - 256
+    - 256
+    output_size: ${eval:'${n_action_steps}*${action_dim}'}
+    activation: relu
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+  obs_dim: ${obs_dim}
+  action_dim: ${action_dim}
+  llm_discriminator:
+    _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+    task_id: ${task_name}
+    llm_translator:
+      _target_: llmbc.translator.llm_translator.LLMTranslator
+      cfg: ${llm}
+      obs_dim: ${task.obs_dim}
+      action_dim: ${task.action_dim}
+      horizon: ${horizon}
+      n_obs_steps: ${n_obs_steps}
+      n_action_steps: ${n_action_steps}
+  loss_bc_weight: 1.0
+  loss_llm_weight: 0.001
+  horizon: ${horizon}
+  n_obs_steps: ${n_obs_steps}
+  n_action_steps: ${n_action_steps}
+  normalize_llm_loss: true
+dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+val_dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 0.01
+  betas:
+  - 0.95
+  - 0.999
+  eps: 1.0e-08
+  weight_decay: 1.0e-06
+training:
+  device: cuda:0
+  seed: 43
+  debug: false
+  resume: false
+  lr_scheduler: cosine
+  lr_warmup_steps: 10
+  num_epochs: 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
+  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-human.pt
+    data_path2: datasets/box-close-v2-human.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: true
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: ete-finetuned
+  finetune_mode: orig
+  checkpoint: data/outputs/2026.05.04/12.16.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-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.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/.hydra/hydra.yaml

@@ -0,0 +1,156 @@
+hydra:
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - training.seed=43
+    - policy.loss_llm_weight=1.0e-3
+  job:
+    name: train
+    chdir: null
+    override_dirname: policy.loss_llm_weight=1.0e-3,training.seed=43
+    id: ???
+    num: ???
+    config_name: llmbc_box-close-v2_human.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /workspace/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /workspace/LLM-BC/config/lang_source
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/LLM-BC/data/outputs/2026.05.05/08.48.27_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

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/.hydra/overrides.yaml

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

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0295-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:104f0809a3395b3cbfb55ca8cf4c229d6f2771cff0e45a02496e40d622fb88c1
+size 864520

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0805-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e291cacaabe98761b0a5cc215a042abeb4dac3474676cffb033319e40691817d
+size 864520

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0890-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7ed8453783072803d452535843a1582accecd956cb5fa64597b119ed9a9c67e
+size 864520

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0925-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ae406bc94176b2e2c11b21e1b4b13a01f9078e34126368256c98540fbe98709
+size 864520

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/checkpoints/epoch=0990-test_success_rate=0.940.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81dfa858ba140b38f1c3228eeb789723c61a2179a74261d2a458d5512ead8c95
+size 864520

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a94470d52f51ef2149f394dd5493beff2a10acffcd7e1a2ed3bf62d98328434
+size 864520

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89cc3e71b35b139478b71ce611e863c49cc61fe6144009ece168b4636d177879
+size 51966530

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/train.log

@@ -0,0 +1,10 @@
+[2026-05-05 08:48:28,824][numexpr.utils][INFO] - Note: NumExpr detected 36 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2026-05-05 08:48:28,824][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2026-05-05 08:48:32,466][datasets][INFO] - PyTorch version 2.2.2 available.
+[2026-05-05 08:48:32,467][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2026-05-05 08:48:32,468][datasets][INFO] - JAX version 0.4.30 available.
+[2026-05-05 08:48:39,277][root][INFO] - running build_ext
+[2026-05-05 08:48:39,280][root][INFO] - building 'mujoco_py.cymj' extension
+[2026-05-05 08:48:39,280][root][INFO] - gcc -pthread -B /venv/llm-bc/compiler_compat -Wno-unused-result -Wsign-compare -DNDEBUG -O2 -Wall -fPIC -O2 -isystem /venv/llm-bc/include -I/venv/llm-bc/include -fPIC -O2 -isystem /venv/llm-bc/include -fPIC -I/venv/llm-bc/lib/python3.9/site-packages/mujoco_py -I/root/.mujoco/mujoco210/include -I/venv/llm-bc/lib/python3.9/site-packages/numpy/core/include -I/venv/llm-bc/include/python3.9 -c /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/cymj.c -o /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/temp.linux-x86_64-cpython-39/venv/llm-bc/lib/python3.9/site-packages/mujoco_py/cymj.o -fopenmp -w
+[2026-05-05 08:49:07,128][root][INFO] - gcc -pthread -B /venv/llm-bc/compiler_compat -Wno-unused-result -Wsign-compare -DNDEBUG -O2 -Wall -fPIC -O2 -isystem /venv/llm-bc/include -I/venv/llm-bc/include -fPIC -O2 -isystem /venv/llm-bc/include -fPIC -I/venv/llm-bc/lib/python3.9/site-packages/mujoco_py -I/root/.mujoco/mujoco210/include -I/venv/llm-bc/lib/python3.9/site-packages/numpy/core/include -I/venv/llm-bc/include/python3.9 -c /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/gl/osmesashim.c -o /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/temp.linux-x86_64-cpython-39/venv/llm-bc/lib/python3.9/site-packages/mujoco_py/gl/osmesashim.o -fopenmp -w
+[2026-05-05 08:49:07,248][root][INFO] - gcc -pthread -B /venv/llm-bc/compiler_compat -shared -Wl,-rpath,/venv/llm-bc/lib -Wl,-rpath-link,/venv/llm-bc/lib -L/venv/llm-bc/lib -L/venv/llm-bc/lib -Wl,-rpath,/venv/llm-bc/lib -Wl,-rpath-link,/venv/llm-bc/lib -L/venv/llm-bc/lib /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/temp.linux-x86_64-cpython-39/venv/llm-bc/lib/python3.9/site-packages/mujoco_py/cymj.o /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/temp.linux-x86_64-cpython-39/venv/llm-bc/lib/python3.9/site-packages/mujoco_py/gl/osmesashim.o -L/root/.mujoco/mujoco210/bin -Wl,--enable-new-dtags,-rpath,/root/.mujoco/mujoco210/bin -lmujoco210 -lglewosmesa -lOSMesa -lGL -o /venv/llm-bc/lib/python3.9/site-packages/mujoco_py/generated/_pyxbld_2.1.2.14_39_linuxcpuextensionbuilder/lib.linux-x86_64-cpython-39/mujoco_py/cymj.cpython-39-x86_64-linux-gnu.so -fopenmp

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_084911-zvpenr2j/files/config.yaml

@@ -0,0 +1,271 @@
+_target_:
+    value: llmbc.workspace.train_llmbc_lowdim_workspace.TrainLLMBCLowdimWorkspace
+_wandb:
+    value:
+        cli_version: 0.18.6
+        m: []
+        python_version: 3.9.25
+        t:
+            "1":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 11
+                - 12
+                - 41
+                - 49
+                - 50
+                - 51
+                - 53
+                - 55
+                - 71
+                - 83
+                - 95
+                - 98
+                - 100
+                - 105
+            "2":
+                - 1
+                - 2
+                - 3
+                - 5
+                - 11
+                - 12
+                - 41
+                - 49
+                - 50
+                - 51
+                - 53
+                - 55
+                - 71
+                - 83
+                - 95
+                - 98
+                - 100
+                - 105
+            "3":
+                - 13
+                - 15
+                - 16
+                - 23
+                - 55
+                - 61
+            "4": 3.9.25
+            "5": 0.18.6
+            "6": 4.47.1
+            "8":
+                - 5
+            "12": 0.18.6
+            "13": linux-x86_64
+action_dim:
+    value: 4
+checkpoint:
+    value:
+        save_last_ckpt: true
+        save_last_snapshot: false
+        topk:
+            format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+            k: 5
+            mode: max
+            monitor_key: test_success_rate
+dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: true
+exp_name:
+    value: default
+horizon:
+    value: 1
+llm:
+    value:
+        causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+        checkpoint: data/outputs/2026.05.04/12.16.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/checkpoint-5890
+        config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+            run:
+                dir: data/outputs/2026.05.05/08.48.27_HuggingFaceTB/SmolLM2-135M-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: SmolLM2-135M-Instruct
+        name: HuggingFaceTB/SmolLM2-135M-Instruct
+        prompter:
+            _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+            use_joint_mlp_projector: true
+        use_joint_mlp_projector: true
+        use_quantization: false
+llm_do_sample:
+    value: false
+llm_orig_expert_feedback:
+    value: true
+logging:
+    value:
+        group: null
+        id: null
+        mode: online
+        name: 2026.05.05-08.48.27_train_llmbc_lowdim_box-close-v2
+        project: box-close-v2-training
+        resume: true
+        tags:
+            - train_llmbc_lowdim
+            - box-close-v2
+            - default
+model_name:
+    value: HuggingFaceTB/SmolLM2-135M-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2
+        wandb_name_base: 2026.05.05-08.48.27_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: /workspace/LLM-BC/data/outputs/2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2
+past_action_visible:
+    value: false
+policy:
+    value:
+        _target_: llmbc.policy.llmbc_lowdim_policy.LLMBCLowdimPolicy
+        action_dim: 4
+        horizon: 1
+        llm_discriminator:
+            _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+            llm_translator:
+                _target_: llmbc.translator.llm_translator.LLMTranslator
+                action_dim: 4
+                cfg:
+                    causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+                    checkpoint: data/outputs/2026.05.04/12.16.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/checkpoint-5890
+                    config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+                        run:
+                            dir: data/outputs/2026.05.05/08.48.27_HuggingFaceTB/SmolLM2-135M-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: SmolLM2-135M-Instruct
+                    name: HuggingFaceTB/SmolLM2-135M-Instruct
+                    prompter:
+                        _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+                        use_joint_mlp_projector: true
+                    use_joint_mlp_projector: true
+                    use_quantization: false
+                horizon: 1
+                n_action_steps: 1
+                n_obs_steps: 1
+                obs_dim: 9
+            task_id: box-close-v2
+        loss_bc_weight: 1
+        loss_llm_weight: 0.001
+        model:
+            _target_: llmbc.model.policy.policy_mlp.PolicyMLP
+            activation: relu
+            hidden_size:
+                - 256
+                - 256
+            input_size: 9
+            n_action_steps: 1
+            n_obs_steps: 1
+            output_size: 4
+        n_action_steps: 1
+        n_obs_steps: 1
+        normalize_llm_loss: true
+        obs_dim: 9
+task:
+    value:
+        action_dim: 4
+        dataset:
+            _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+            data_path: datasets/box-close-v2-human.pt
+            data_path2: datasets/box-close-v2-human.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: 43
+        tqdm_interval_sec: 1
+        val_every: 1
+val_dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: true

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_084911-zvpenr2j/files/output.log

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

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_084911-zvpenr2j/files/wandb-metadata.json

@@ -0,0 +1,57 @@
+{
+  "os": "Linux-5.15.0-140-generic-x86_64-with-glibc2.35",
+  "python": "3.9.25",
+  "startedAt": "2026-05-05T08:49:11.307995Z",
+  "args": [
+    "--config-path",
+    "config/lang_source",
+    "--config-name",
+    "llmbc_box-close-v2_human.yaml",
+    "training.seed=43",
+    "policy.loss_llm_weight=1.0e-3"
+  ],
+  "program": "/workspace/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "289618e617b0141d7a8a634ebe196cba52d07960"
+  },
+  "email": "chris920325@gmail.com",
+  "root": "/workspace/LLM-BC/data/outputs/2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2",
+  "host": "d8640274e028",
+  "username": "root",
+  "executable": "/venv/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 36,
+  "cpu_count_logical": 36,
+  "gpu": "NVIDIA GeForce RTX 4090",
+  "gpu_count": 2,
+  "disk": {
+    "/": {
+      "total": "569083166720",
+      "used": "425902718976"
+    }
+  },
+  "memory": {
+    "total": "135036993536"
+  },
+  "cpu": {
+    "count": 36,
+    "countLogical": 36
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA GeForce RTX 4090",
+      "memoryTotal": "25757220864",
+      "cudaCores": 16384,
+      "architecture": "Ada"
+    },
+    {
+      "name": "NVIDIA GeForce RTX 4090",
+      "memoryTotal": "25757220864",
+      "cudaCores": 16384,
+      "architecture": "Ada"
+    }
+  ],
+  "cudaVersion": "12.4"
+}

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_084911-zvpenr2j/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_timestamp":1.778018440130274e+09,"_step":238237,"train_loss_llm":0.0009095072746276855,"train/mean_score":7.897461311760635,"global_step":238237,"lr":2.7842694261259738e-11,"train/cumulative_reward":16.01912307196755,"_wandb":{"runtime":47488},"grad_norm":0.00290088658221066,"train_action_mse_error":1.0189558452111669e-05,"test/mean_score":7.445443287977502,"_runtime":47488.847103069,"test/cumulative_reward":13.693659910308838,"val_loss":0.0007264076848514378,"train/success_rate":1,"epoch":1000,"train_loss_bc":2.5763029043446295e-06,"test/success_rate":0.86,"train_loss":0.00018306689496072802}

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/wandb/run-20260505_084911-zvpenr2j/run-zvpenr2j.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1fb51d725547b41489adc96aff24580238cc030b9a96528dcc2c83c14adbfb5
+size 197966535

2026.05.05/08.48.27_train_llmbc_lowdim_box-close-v2/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "zvpenr2j"}