Upload folder using huggingface_hub

.gitattributes

@@ -470,3 +470,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 2026.04.18/08.56.49_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260418_085750-q6l6s41x/run-q6l6s41x.wandb filter=lfs diff=lfs merge=lfs -text
 2026.04.18/09.06.44_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/logs.json.txt filter=lfs diff=lfs merge=lfs -text
 2026.04.18/09.06.44_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260418_090740-6g56arly/run-6g56arly.wandb filter=lfs diff=lfs merge=lfs -text
+2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164316-n0n0l1oe/run-n0n0l1oe.wandb filter=lfs diff=lfs merge=lfs -text
+2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164318-p0jndf7b/run-p0jndf7b.wandb filter=lfs diff=lfs merge=lfs -text

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/.hydra/config.yaml

@@ -0,0 +1,194 @@
+name: train_llm_diffusion_unet_lowdim
+_target_: llmbc.workspace.train_llm_diffusion_unet_lowdim_workspace.TrainLLMDiffusionUnetLowdimWorkspace
+obs_dim: ${task.obs_dim}
+action_dim: ${task.action_dim}
+task_name: ${task.name}
+exp_name: default
+model_name: ${llm.name}
+horizon: 1
+n_obs_steps: 1
+n_action_steps: 1
+n_latency_steps: 0
+past_action_visible: false
+keypoint_visible_rate: 1.0
+obs_as_local_cond: false
+obs_as_global_cond: true
+pred_action_steps_only: false
+llm_orig_expert_feedback: true
+llm_do_sample: false
+policy:
+  _target_: llmbc.policy.llm_diffusion_unet_lowdim_policy.LLMDiffusionUnetLowdimPolicy
+  model:
+    _target_: llmbc.model.diffusion.conditional_unet1d.ConditionalUnet1D
+    input_dim: '${eval: ${task.action_dim} if ${obs_as_local_cond} or ${obs_as_global_cond}
+      else ${task.obs_dim} + ${task.action_dim}}'
+    local_cond_dim: '${eval: ${task.obs_dim} if ${obs_as_local_cond} else None}'
+    global_cond_dim: '${eval: ${task.obs_dim}*${n_obs_steps} if ${obs_as_global_cond}
+      else None}'
+    diffusion_step_embed_dim: 256
+    down_dims:
+    - 256
+    - 512
+    - 1024
+    kernel_size: 5
+    n_groups: 8
+    cond_predict_scale: true
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddpm.DDPMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    beta_schedule: squaredcos_cap_v2
+    variance_type: fixed_small
+    clip_sample: true
+    prediction_type: epsilon
+  horizon: ${horizon}
+  obs_dim: ${obs_dim}
+  action_dim: ${action_dim}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_local_cond: ${obs_as_local_cond}
+  obs_as_global_cond: ${obs_as_global_cond}
+  pred_action_steps_only: ${pred_action_steps_only}
+  oa_step_convention: true
+  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: ${obs_dim}
+      action_dim: ${action_dim}
+      horizon: ${horizon}
+      n_obs_steps: ${n_obs_steps}
+      n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  loss_dp_weight: 1.0
+  loss_llm_weight: 0.001
+  normalize_llm_loss: true
+  norm_batch: true
+  reweight_llm_loss: true
+ema:
+  _target_: llmbc.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+val_dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: false
+  pin_memory: false
+  persistent_workers: false
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 0.0001
+  betas:
+  - 0.95
+  - 0.999
+  eps: 1.0e-08
+  weight_decay: 1.0e-06
+training:
+  device: cuda:0
+  seed: 43
+  debug: false
+  resume: true
+  lr_scheduler: cosine
+  lr_warmup_steps: 500
+  num_epochs: 1001
+  gradient_accumulate_every: 16
+  use_ema: true
+  rollout_every: 5
+  checkpoint_every: 5
+  val_every: 1
+  sample_every: 5
+  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: adroit-hand-relocate-v1
+  obs_dim: 39
+  action_dim: 30
+  env_runner:
+    _target_: llmbc.env_runner.adroit_lowdim_runner.AdroitHandLowdimRunner
+    env_name: llf-adroit-adroit-hand-relocate-v1
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 150
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.99
+  dataset:
+    _target_: llmbc.dataset.adroit_lowdim_dataset.AdroitHandLowdimDataset
+    data_path: datasets/adroit-hand-relocate-v1.pt
+    data_path2: datasets/adroit-hand-relocate-v1.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: false
+  instructor:
+    _target_: llmbc.translator.instructor.adroit_instructor.adroit_hand_relocate_v1_instructor.AdroitHandRelocateV1Instructor
+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/2026.04.03/12.44.33_train_llm_lowdim_adroit-hand-relocate-v1/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-adroit-hand-relocate-v1/checkpoint-2340
+  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}

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/.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:
+    - training.seed=43
+  job:
+    name: train
+    chdir: null
+    override_dirname: training.seed=43
+    id: ???
+    num: ???
+    config_name: llmdp_adroit-hand-relocate-v1.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/main_table
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/LLM-BC/data/outputs/2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+    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

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- training.seed=43

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0615-test_success_rate=0.800.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5522081ed5ebceb04d3802db39f615f9c4c6be6a603f93e78b4aba2049eef4e
+size 1048128562

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0620-test_success_rate=0.720.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:566a6644da07e556b85b73b08a4518b86d4d5c2db0def23744d95daa95ec0c24
+size 1048128562

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0710-test_success_rate=0.720.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91e75fddadbaea34bb99b21282746fcabde3976729bed2d2fd4dfa61776e6389
+size 1048128562

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0830-test_success_rate=0.740.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e619db0b15b7e6831caebe75d083b4fb7df870379b38a6e51609db94b14926b
+size 1048128562

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0880-test_success_rate=0.740.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2406bc138d04203d667057bf40cd9ae42044dd50f7a0565e84db317a2db6fda
+size 1048128562

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:369d53d375d4e24c6680ab2d934d60db244cfd6ff92c634d59613b20f8366ce7
+size 1048128562

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df5319410741cda2ffb19bacc4ab6d31783db227b96936a58ae254d9d8dac2ad
+size 10766179

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/train.log

@@ -0,0 +1,14 @@
+[2026-04-19 16:41:40,721][numexpr.utils][INFO] - Note: detected 144 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable.
+[2026-04-19 16:41:40,721][numexpr.utils][INFO] - Note: NumExpr detected 144 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2026-04-19 16:41:40,721][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2026-04-19 16:41:42,919][datasets][INFO] - PyTorch version 2.2.2 available.
+[2026-04-19 16:41:42,919][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2026-04-19 16:41:42,920][datasets][INFO] - JAX version 0.4.30 available.
+[2026-04-19 16:41:44,603][llmbc.model.diffusion.conditional_unet1d][INFO] - number of parameters: 6.549149e+07
+[2026-04-19 16:41:49,205][root][INFO] - running build_ext
+[2026-04-19 16:41:49,207][root][INFO] - building 'mujoco_py.cymj' extension
+[2026-04-19 16:41:49,207][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
+[2026-04-19 16:42:31,763][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
+[2026-04-19 16:42:31,908][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
+[2026-04-19 16:43:15,462][absl][INFO] - MUJOCO_GL=osmesa, attempting to import specified OpenGL backend.
+[2026-04-19 16:43:15,471][absl][INFO] - MuJoCo library version is: 2.3.7

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164316-n0n0l1oe/files/config.yaml

@@ -0,0 +1,305 @@
+_target_:
+    value: llmbc.workspace.train_llm_diffusion_unet_lowdim_workspace.TrainLLMDiffusionUnetLowdimWorkspace
+_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: 30
+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
+ema:
+    value:
+        _target_: llmbc.model.diffusion.ema_model.EMAModel
+        inv_gamma: 1
+        max_value: 0.9999
+        min_value: 0
+        power: 0.75
+        update_after_step: 0
+exp_name:
+    value: default
+horizon:
+    value: 1
+keypoint_visible_rate:
+    value: 1
+llm:
+    value:
+        causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+        checkpoint: data/outputs/2026.04.03/12.44.33_train_llm_lowdim_adroit-hand-relocate-v1/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-adroit-hand-relocate-v1/checkpoint-2340
+        config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+            run:
+                dir: data/outputs/2026.04.19/16.41.39_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: 2026.04.19-16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+        project: adroit-hand-relocate-v1-training
+        resume: true
+        tags:
+            - train_llm_diffusion_unet_lowdim
+            - adroit-hand-relocate-v1
+            - default
+model_name:
+    value: HuggingFaceTB/SmolLM2-135M-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+        wandb_name_base: 2026.04.19-16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+n_action_steps:
+    value: 1
+n_latency_steps:
+    value: 0
+n_obs_steps:
+    value: 1
+name:
+    value: train_llm_diffusion_unet_lowdim
+obs_as_global_cond:
+    value: true
+obs_as_local_cond:
+    value: false
+obs_dim:
+    value: 39
+optimizer:
+    value:
+        _target_: torch.optim.AdamW
+        betas:
+            - 0.95
+            - 0.999
+        eps: 1e-08
+        lr: 0.0001
+        weight_decay: 1e-06
+output_dir:
+    value: /workspace/LLM-BC/data/outputs/2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+past_action_visible:
+    value: false
+policy:
+    value:
+        _target_: llmbc.policy.llm_diffusion_unet_lowdim_policy.LLMDiffusionUnetLowdimPolicy
+        action_dim: 30
+        horizon: 1
+        llm_discriminator:
+            _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+            llm_translator:
+                _target_: llmbc.translator.llm_translator.LLMTranslator
+                action_dim: 30
+                cfg:
+                    causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+                    checkpoint: data/outputs/2026.04.03/12.44.33_train_llm_lowdim_adroit-hand-relocate-v1/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-adroit-hand-relocate-v1/checkpoint-2340
+                    config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+                        run:
+                            dir: data/outputs/2026.04.19/16.41.39_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: 39
+            task_id: adroit-hand-relocate-v1
+        loss_dp_weight: 1
+        loss_llm_weight: 0.001
+        model:
+            _target_: llmbc.model.diffusion.conditional_unet1d.ConditionalUnet1D
+            cond_predict_scale: true
+            diffusion_step_embed_dim: 256
+            down_dims:
+                - 256
+                - 512
+                - 1024
+            global_cond_dim: 39
+            input_dim: 30
+            kernel_size: 5
+            local_cond_dim: null
+            n_groups: 8
+        n_action_steps: 1
+        n_obs_steps: 1
+        noise_scheduler:
+            _target_: diffusers.schedulers.scheduling_ddpm.DDPMScheduler
+            beta_end: 0.02
+            beta_schedule: squaredcos_cap_v2
+            beta_start: 0.0001
+            clip_sample: true
+            num_train_timesteps: 100
+            prediction_type: epsilon
+            variance_type: fixed_small
+        norm_batch: true
+        normalize_llm_loss: true
+        num_inference_steps: 100
+        oa_step_convention: true
+        obs_as_global_cond: true
+        obs_as_local_cond: false
+        obs_dim: 39
+        pred_action_steps_only: false
+        reweight_llm_loss: true
+pred_action_steps_only:
+    value: false
+task:
+    value:
+        action_dim: 30
+        dataset:
+            _target_: llmbc.dataset.adroit_lowdim_dataset.AdroitHandLowdimDataset
+            data_path: datasets/adroit-hand-relocate-v1.pt
+            data_path2: datasets/adroit-hand-relocate-v1.pt
+            dummy_normalizer: false
+            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.adroit_lowdim_runner.AdroitHandLowdimRunner
+            discount: 0.99
+            env_name: llf-adroit-adroit-hand-relocate-v1
+            feedback_type:
+                - hp
+                - hn
+                - fp
+            instruction_type: b
+            max_steps: 150
+            n_action_steps: 1
+            n_envs: 10
+            n_obs_steps: 1
+            n_test: 50
+            n_train: 10
+            visual: false
+        instructor:
+            _target_: llmbc.translator.instructor.adroit_instructor.adroit_hand_relocate_v1_instructor.AdroitHandRelocateV1Instructor
+        name: adroit-hand-relocate-v1
+        obs_dim: 39
+task_name:
+    value: adroit-hand-relocate-v1
+training:
+    value:
+        checkpoint_every: 5
+        debug: false
+        device: cuda:0
+        gradient_accumulate_every: 16
+        lr_scheduler: cosine
+        lr_warmup_steps: 500
+        max_train_steps: null
+        max_val_steps: null
+        num_epochs: 1001
+        resume: true
+        rollout_every: 5
+        sample_every: 5
+        seed: 43
+        tqdm_interval_sec: 1
+        use_ema: true
+        val_every: 1
+val_dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: false

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164316-n0n0l1oe/files/output.log

@@ -0,0 +1,3 @@
+Eval AdroitHandLowdimRunner 1/6:   0%|                                                                                                                     | 0/150 [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)
+                                                                                                                                                                                   

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164316-n0n0l1oe/files/wandb-metadata.json

@@ -0,0 +1,56 @@
+{
+  "os": "Linux-5.15.0-168-generic-x86_64-with-glibc2.39",
+  "python": "3.9.25",
+  "startedAt": "2026-04-19T16:43:16.534883Z",
+  "args": [
+    "--config-path",
+    "config/main_table",
+    "--config-name",
+    "llmdp_adroit-hand-relocate-v1.yaml",
+    "training.seed=43"
+  ],
+  "program": "/workspace/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "e628f36961d2a26f3afecf72e98540797ae35384"
+  },
+  "email": "chris920325@gmail.com",
+  "root": "/workspace/LLM-BC/data/outputs/2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1",
+  "host": "c8ef52649900",
+  "username": "root",
+  "executable": "/venv/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 72,
+  "cpu_count_logical": 144,
+  "gpu": "NVIDIA GeForce RTX 4090",
+  "gpu_count": 2,
+  "disk": {
+    "/": {
+      "total": "545460846592",
+      "used": "402739404800"
+    }
+  },
+  "memory": {
+    "total": "810942898176"
+  },
+  "cpu": {
+    "count": 72,
+    "countLogical": 144
+  },
+  "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": "12.8"
+}

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164316-n0n0l1oe/files/wandb-summary.json

@@ -0,0 +1 @@
+{"test/cumulative_reward":133.4980267839532,"train/success_rate":0.7,"_wandb":{"runtime":127603},"lr":2.0838688302715626e-11,"test/success_rate":0.66,"train_loss_llm":0,"train/cumulative_reward":88.3015347915036,"_step":63062,"global_step":63062,"train_action_mse_error":0.04935038834810257,"train/mean_score":23.926532919958866,"test/mean_score":23.73805529603139,"val_loss":0.42147260904312134,"train_loss_dp":0.050730641931295395,"_runtime":127603.431494976,"train_loss":0.042420177823967405,"epoch":1000,"_timestamp":1.7767445990360188e+09}

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164316-n0n0l1oe/run-n0n0l1oe.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42d3c72e70e6b33345bc667ae2a15e2c50d2b42b71e77df257b2abd0da6098ba
+size 192749139

2026.04.19/16.41.39_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "n0n0l1oe"}

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/.hydra/config.yaml

@@ -0,0 +1,194 @@
+name: train_llm_diffusion_unet_lowdim
+_target_: llmbc.workspace.train_llm_diffusion_unet_lowdim_workspace.TrainLLMDiffusionUnetLowdimWorkspace
+obs_dim: ${task.obs_dim}
+action_dim: ${task.action_dim}
+task_name: ${task.name}
+exp_name: default
+model_name: ${llm.name}
+horizon: 1
+n_obs_steps: 1
+n_action_steps: 1
+n_latency_steps: 0
+past_action_visible: false
+keypoint_visible_rate: 1.0
+obs_as_local_cond: false
+obs_as_global_cond: true
+pred_action_steps_only: false
+llm_orig_expert_feedback: true
+llm_do_sample: false
+policy:
+  _target_: llmbc.policy.llm_diffusion_unet_lowdim_policy.LLMDiffusionUnetLowdimPolicy
+  model:
+    _target_: llmbc.model.diffusion.conditional_unet1d.ConditionalUnet1D
+    input_dim: '${eval: ${task.action_dim} if ${obs_as_local_cond} or ${obs_as_global_cond}
+      else ${task.obs_dim} + ${task.action_dim}}'
+    local_cond_dim: '${eval: ${task.obs_dim} if ${obs_as_local_cond} else None}'
+    global_cond_dim: '${eval: ${task.obs_dim}*${n_obs_steps} if ${obs_as_global_cond}
+      else None}'
+    diffusion_step_embed_dim: 256
+    down_dims:
+    - 256
+    - 512
+    - 1024
+    kernel_size: 5
+    n_groups: 8
+    cond_predict_scale: true
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddpm.DDPMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    beta_schedule: squaredcos_cap_v2
+    variance_type: fixed_small
+    clip_sample: true
+    prediction_type: epsilon
+  horizon: ${horizon}
+  obs_dim: ${obs_dim}
+  action_dim: ${action_dim}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_local_cond: ${obs_as_local_cond}
+  obs_as_global_cond: ${obs_as_global_cond}
+  pred_action_steps_only: ${pred_action_steps_only}
+  oa_step_convention: true
+  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: ${obs_dim}
+      action_dim: ${action_dim}
+      horizon: ${horizon}
+      n_obs_steps: ${n_obs_steps}
+      n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  loss_dp_weight: 1.0
+  loss_llm_weight: 0.001
+  normalize_llm_loss: true
+  norm_batch: true
+  reweight_llm_loss: true
+ema:
+  _target_: llmbc.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: true
+  pin_memory: false
+  persistent_workers: false
+val_dataloader:
+  batch_size: 16
+  num_workers: 0
+  shuffle: false
+  pin_memory: false
+  persistent_workers: false
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 0.0001
+  betas:
+  - 0.95
+  - 0.999
+  eps: 1.0e-08
+  weight_decay: 1.0e-06
+training:
+  device: cuda:0
+  seed: 44
+  debug: false
+  resume: true
+  lr_scheduler: cosine
+  lr_warmup_steps: 500
+  num_epochs: 1001
+  gradient_accumulate_every: 16
+  use_ema: true
+  rollout_every: 5
+  checkpoint_every: 5
+  val_every: 1
+  sample_every: 5
+  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: adroit-hand-relocate-v1
+  obs_dim: 39
+  action_dim: 30
+  env_runner:
+    _target_: llmbc.env_runner.adroit_lowdim_runner.AdroitHandLowdimRunner
+    env_name: llf-adroit-adroit-hand-relocate-v1
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 150
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.99
+  dataset:
+    _target_: llmbc.dataset.adroit_lowdim_dataset.AdroitHandLowdimDataset
+    data_path: datasets/adroit-hand-relocate-v1.pt
+    data_path2: datasets/adroit-hand-relocate-v1.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: false
+  instructor:
+    _target_: llmbc.translator.instructor.adroit_instructor.adroit_hand_relocate_v1_instructor.AdroitHandRelocateV1Instructor
+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/2026.04.03/12.44.33_train_llm_lowdim_adroit-hand-relocate-v1/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-adroit-hand-relocate-v1/checkpoint-2340
+  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}

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/.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:
+    - training.seed=44
+  job:
+    name: train
+    chdir: null
+    override_dirname: training.seed=44
+    id: ???
+    num: ???
+    config_name: llmdp_adroit-hand-relocate-v1.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/main_table
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/LLM-BC/data/outputs/2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+    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

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- training.seed=44

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0610-test_success_rate=0.740.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:505eaf2c73ed51a913a62a3734f452acd67879d81f8accb6838e6d62c832c890
+size 1048128562

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0705-test_success_rate=0.740.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74fe2ef3507b3e02c6cf82067797f129462f4506238e259556ce52877e4566ae
+size 1048128562

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0745-test_success_rate=0.760.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cca810176a6ae4a31943c1c855e78136dabf6263ad6a415fd7efecd5e11c982b
+size 1048128562

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0855-test_success_rate=0.760.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f8b23cff4e1bae521753e10eea1d244f4acdbdce0a8a22b95e9d0763b8ca163
+size 1048128562

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/epoch=0890-test_success_rate=0.760.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6b04e55288482354a3c41c5ba9aff2f6aa49c72b26893a448ea66434928b4b6
+size 1048128562

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2144f314adbbb8b74d6aef8124c31f98a65405f514efedcc7920c77a4e2b37f7
+size 1048128562

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d952989144d447dfd4b00a27b69f084a731e8c47ab23eef80b77ac153b32a703
+size 10770177

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/train.log

@@ -0,0 +1,14 @@
+[2026-04-19 16:41:54,634][numexpr.utils][INFO] - Note: detected 144 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable.
+[2026-04-19 16:41:54,635][numexpr.utils][INFO] - Note: NumExpr detected 144 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2026-04-19 16:41:54,635][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2026-04-19 16:41:56,862][datasets][INFO] - PyTorch version 2.2.2 available.
+[2026-04-19 16:41:56,862][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2026-04-19 16:41:56,863][datasets][INFO] - JAX version 0.4.30 available.
+[2026-04-19 16:41:58,451][llmbc.model.diffusion.conditional_unet1d][INFO] - number of parameters: 6.549149e+07
+[2026-04-19 16:42:32,648][root][INFO] - running build_ext
+[2026-04-19 16:42:32,651][root][INFO] - building 'mujoco_py.cymj' extension
+[2026-04-19 16:42:32,652][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
+[2026-04-19 16:43:14,361][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
+[2026-04-19 16:43:14,476][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
+[2026-04-19 16:43:17,318][absl][INFO] - MUJOCO_GL=osmesa, attempting to import specified OpenGL backend.
+[2026-04-19 16:43:17,324][absl][INFO] - MuJoCo library version is: 2.3.7

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164318-p0jndf7b/files/config.yaml

@@ -0,0 +1,305 @@
+_target_:
+    value: llmbc.workspace.train_llm_diffusion_unet_lowdim_workspace.TrainLLMDiffusionUnetLowdimWorkspace
+_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: 30
+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
+ema:
+    value:
+        _target_: llmbc.model.diffusion.ema_model.EMAModel
+        inv_gamma: 1
+        max_value: 0.9999
+        min_value: 0
+        power: 0.75
+        update_after_step: 0
+exp_name:
+    value: default
+horizon:
+    value: 1
+keypoint_visible_rate:
+    value: 1
+llm:
+    value:
+        causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+        checkpoint: data/outputs/2026.04.03/12.44.33_train_llm_lowdim_adroit-hand-relocate-v1/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-adroit-hand-relocate-v1/checkpoint-2340
+        config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+            run:
+                dir: data/outputs/2026.04.19/16.41.52_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: 2026.04.19-16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+        project: adroit-hand-relocate-v1-training
+        resume: true
+        tags:
+            - train_llm_diffusion_unet_lowdim
+            - adroit-hand-relocate-v1
+            - default
+model_name:
+    value: HuggingFaceTB/SmolLM2-135M-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+        wandb_name_base: 2026.04.19-16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+n_action_steps:
+    value: 1
+n_latency_steps:
+    value: 0
+n_obs_steps:
+    value: 1
+name:
+    value: train_llm_diffusion_unet_lowdim
+obs_as_global_cond:
+    value: true
+obs_as_local_cond:
+    value: false
+obs_dim:
+    value: 39
+optimizer:
+    value:
+        _target_: torch.optim.AdamW
+        betas:
+            - 0.95
+            - 0.999
+        eps: 1e-08
+        lr: 0.0001
+        weight_decay: 1e-06
+output_dir:
+    value: /workspace/LLM-BC/data/outputs/2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1
+past_action_visible:
+    value: false
+policy:
+    value:
+        _target_: llmbc.policy.llm_diffusion_unet_lowdim_policy.LLMDiffusionUnetLowdimPolicy
+        action_dim: 30
+        horizon: 1
+        llm_discriminator:
+            _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+            llm_translator:
+                _target_: llmbc.translator.llm_translator.LLMTranslator
+                action_dim: 30
+                cfg:
+                    causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+                    checkpoint: data/outputs/2026.04.03/12.44.33_train_llm_lowdim_adroit-hand-relocate-v1/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-adroit-hand-relocate-v1/checkpoint-2340
+                    config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+                        run:
+                            dir: data/outputs/2026.04.19/16.41.52_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: 39
+            task_id: adroit-hand-relocate-v1
+        loss_dp_weight: 1
+        loss_llm_weight: 0.001
+        model:
+            _target_: llmbc.model.diffusion.conditional_unet1d.ConditionalUnet1D
+            cond_predict_scale: true
+            diffusion_step_embed_dim: 256
+            down_dims:
+                - 256
+                - 512
+                - 1024
+            global_cond_dim: 39
+            input_dim: 30
+            kernel_size: 5
+            local_cond_dim: null
+            n_groups: 8
+        n_action_steps: 1
+        n_obs_steps: 1
+        noise_scheduler:
+            _target_: diffusers.schedulers.scheduling_ddpm.DDPMScheduler
+            beta_end: 0.02
+            beta_schedule: squaredcos_cap_v2
+            beta_start: 0.0001
+            clip_sample: true
+            num_train_timesteps: 100
+            prediction_type: epsilon
+            variance_type: fixed_small
+        norm_batch: true
+        normalize_llm_loss: true
+        num_inference_steps: 100
+        oa_step_convention: true
+        obs_as_global_cond: true
+        obs_as_local_cond: false
+        obs_dim: 39
+        pred_action_steps_only: false
+        reweight_llm_loss: true
+pred_action_steps_only:
+    value: false
+task:
+    value:
+        action_dim: 30
+        dataset:
+            _target_: llmbc.dataset.adroit_lowdim_dataset.AdroitHandLowdimDataset
+            data_path: datasets/adroit-hand-relocate-v1.pt
+            data_path2: datasets/adroit-hand-relocate-v1.pt
+            dummy_normalizer: false
+            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.adroit_lowdim_runner.AdroitHandLowdimRunner
+            discount: 0.99
+            env_name: llf-adroit-adroit-hand-relocate-v1
+            feedback_type:
+                - hp
+                - hn
+                - fp
+            instruction_type: b
+            max_steps: 150
+            n_action_steps: 1
+            n_envs: 10
+            n_obs_steps: 1
+            n_test: 50
+            n_train: 10
+            visual: false
+        instructor:
+            _target_: llmbc.translator.instructor.adroit_instructor.adroit_hand_relocate_v1_instructor.AdroitHandRelocateV1Instructor
+        name: adroit-hand-relocate-v1
+        obs_dim: 39
+task_name:
+    value: adroit-hand-relocate-v1
+training:
+    value:
+        checkpoint_every: 5
+        debug: false
+        device: cuda:0
+        gradient_accumulate_every: 16
+        lr_scheduler: cosine
+        lr_warmup_steps: 500
+        max_train_steps: null
+        max_val_steps: null
+        num_epochs: 1001
+        resume: true
+        rollout_every: 5
+        sample_every: 5
+        seed: 44
+        tqdm_interval_sec: 1
+        use_ema: true
+        val_every: 1
+val_dataloader:
+    value:
+        batch_size: 16
+        num_workers: 0
+        persistent_workers: false
+        pin_memory: false
+        shuffle: false

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164318-p0jndf7b/files/output.log

@@ -0,0 +1,3 @@
+Eval AdroitHandLowdimRunner 1/6:   0%|                                                                                                                     | 0/150 [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)
+                                                                                                                                                                                   

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164318-p0jndf7b/files/wandb-metadata.json

@@ -0,0 +1,56 @@
+{
+  "os": "Linux-5.15.0-168-generic-x86_64-with-glibc2.39",
+  "python": "3.9.25",
+  "startedAt": "2026-04-19T16:43:18.292265Z",
+  "args": [
+    "--config-path",
+    "config/main_table",
+    "--config-name",
+    "llmdp_adroit-hand-relocate-v1.yaml",
+    "training.seed=44"
+  ],
+  "program": "/workspace/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "e628f36961d2a26f3afecf72e98540797ae35384"
+  },
+  "email": "chris920325@gmail.com",
+  "root": "/workspace/LLM-BC/data/outputs/2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1",
+  "host": "c8ef52649900",
+  "username": "root",
+  "executable": "/venv/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 72,
+  "cpu_count_logical": 144,
+  "gpu": "NVIDIA GeForce RTX 4090",
+  "gpu_count": 2,
+  "disk": {
+    "/": {
+      "total": "545460846592",
+      "used": "402739441664"
+    }
+  },
+  "memory": {
+    "total": "810942898176"
+  },
+  "cpu": {
+    "count": 72,
+    "countLogical": 144
+  },
+  "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": "12.8"
+}

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164318-p0jndf7b/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_runtime":129854.646867075,"_timestamp":1.776746851850192e+09,"train/cumulative_reward":161.88687393688787,"train/mean_score":26.946863930148574,"train_loss":0.04129583752226262,"train_action_mse_error":0.03687829524278641,"test/success_rate":0.66,"val_loss":0.22105738520622253,"lr":2.0838688302715626e-11,"train_loss_llm":0.0013622492551803589,"_step":63062,"epoch":1000,"test/cumulative_reward":129.2000056830647,"global_step":63062,"test/mean_score":23.523855289598842,"train/success_rate":0.8,"_wandb":{"runtime":129854},"train_loss_dp":0.024869214743375778}

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/run-20260419_164318-p0jndf7b/run-p0jndf7b.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c6c9b32353175d7b3461a81a724e85a5f3660066950119944d894c67a4709a2
+size 193142946

2026.04.19/16.41.52_train_llm_diffusion_unet_lowdim_adroit-hand-relocate-v1/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "p0jndf7b"}