Upload folder using huggingface_hub

.gitattributes

@@ -66,3 +66,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 2025.12.20/07.32.01_train_llm_diffusion_unet_lowdim_parking-v0/wandb/run-20251220_073213-wn6bouke/run-wn6bouke.wandb filter=lfs diff=lfs merge=lfs -text
 2025.12.20/07.32.13_train_llm_diffusion_unet_lowdim_parking-v0/logs.json.txt filter=lfs diff=lfs merge=lfs -text
 2025.12.20/07.32.13_train_llm_diffusion_unet_lowdim_parking-v0/wandb/run-20251220_073225-wvteizn9/run-wvteizn9.wandb filter=lfs diff=lfs merge=lfs -text
+2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/run-4uxuujzc.wandb filter=lfs diff=lfs merge=lfs -text
+2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/run-653bfiyg.wandb filter=lfs diff=lfs merge=lfs -text
+2025.12.23/03.21.15_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text
+2025.12.23/03.21.15_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032128-lb3ukhr7/run-lb3ukhr7.wandb filter=lfs diff=lfs merge=lfs -text

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/config.yaml

@@ -0,0 +1,193 @@
+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: 0.01
+  normalize_llm_loss: 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: 42
+  debug: false
+  resume: true
+  lr_scheduler: cosine
+  lr_warmup_steps: 500
+  num_epochs: 20001
+  gradient_accumulate_every: 16
+  use_ema: true
+  rollout_every: 100
+  checkpoint_every: 100
+  val_every: 1
+  sample_every: 100
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-split-ablation
+  resume: true
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags:
+  - ${name}
+  - ${task_name}
+  - ${exp_name}
+  id: null
+  group: null
+checkpoint:
+  topk:
+    monitor_key: test_success_rate
+    mode: max
+    k: 5
+    format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+  save_last_ckpt: true
+  save_last_snapshot: false
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+task:
+  name: box-close-v2
+  obs_dim: 9
+  action_dim: 4
+  env_runner:
+    _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+    env_name: llf-metaworld-box-close-v2
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 30
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.9
+  dataset:
+    _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+    data_path: datasets/box-close-v2-split0.05.pt
+    data_path2: datasets/box-close-v2-split0.05.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: false
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: pretrained
+  finetune_mode: orig
+  checkpoint: data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+  max_length: 100
+  lora_config:
+    r: 32
+    lora_alpha: 64
+    lora_dropout: 0.05
+    bias: none
+    task_type: CAUSAL_LM
+  prompter:
+    _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/hydra.yaml

@@ -0,0 +1,155 @@
+hydra:
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - policy.loss_llm_weight=1.0e-2
+  job:
+    name: train
+    chdir: null
+    override_dirname: policy.loss_llm_weight=1.0e-2
+    id: ???
+    num: ???
+    config_name: llmdp_box-close-v2_pretrained_split0.05.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /root/workspace/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /root/workspace/LLM-BC/config/data_efficiency
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2
+    choices:
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- policy.loss_llm_weight=1.0e-2

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=2000-test_success_rate=0.800.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e144ee2536674a3b417d6fe31b2922245ca29a69b3189688c4bcdac4101790b6
+size 1040499058

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=2500-test_success_rate=0.800.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2d01a32802acb0d3e0855aa39224c5035a91ffedee28a1849e7b00007b29e7c
+size 1040499058

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=2800-test_success_rate=0.820.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18f356123523b91d23bec48724c32814d967f3c60b95a839a490285eace825b3
+size 1040499058

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=3100-test_success_rate=0.840.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fb09873896003faf8b970bf8ba0561e942000635178584dd5c58bf034614602
+size 1040499058

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=3400-test_success_rate=0.800.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b87c0ea26dd51f0c33fc610c481da1f9c2186ab4614d5bfb6d7bb18fcde0c46d
+size 1040499058

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6055bc7e2666ab3619c87c1cba77b1158f10813567db9af5a2c0bda321559a5
+size 1040499058

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70cb7a408cb7fe2b3f8735700e090edd9c54c3965a350ed428462e48d9a6c048
+size 46090340

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/train.log

@@ -0,0 +1,9 @@
+[2025-12-23 03:20:57,683][numexpr.utils][INFO] - Note: detected 128 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable.
+[2025-12-23 03:20:57,683][numexpr.utils][INFO] - Note: NumExpr detected 128 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2025-12-23 03:20:57,683][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2025-12-23 03:20:59,596][datasets][INFO] - PyTorch version 2.2.2 available.
+[2025-12-23 03:20:59,596][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2025-12-23 03:20:59,597][datasets][INFO] - JAX version 0.4.30 available.
+[2025-12-23 03:21:01,220][llmbc.model.diffusion.conditional_unet1d][INFO] - number of parameters: 6.501479e+07
+[2025-12-23 03:21:06,644][absl][INFO] - MUJOCO_GL=osmesa, attempting to import specified OpenGL backend.
+[2025-12-23 03:21:06,649][absl][INFO] - MuJoCo library version is: 2.3.7

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug-internal.log

@@ -0,0 +1,17 @@
+{"time":"2025-12-23T03:21:07.491073659Z","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2025-12-23T03:21:07.491085059Z","level":"INFO","msg":"created symlink","path":"/root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/logs/debug-core.log"}
+{"time":"2025-12-23T03:21:07.600486464Z","level":"INFO","msg":"created new stream","id":"4uxuujzc"}
+{"time":"2025-12-23T03:21:07.600515463Z","level":"INFO","msg":"stream: started","id":"4uxuujzc"}
+{"time":"2025-12-23T03:21:07.600540399Z","level":"INFO","msg":"sender: started","stream_id":"4uxuujzc"}
+{"time":"2025-12-23T03:21:07.600545933Z","level":"INFO","msg":"handler: started","stream_id":{"value":"4uxuujzc"}}
+{"time":"2025-12-23T03:21:07.600531776Z","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"4uxuujzc"}}
+{"time":"2025-12-23T03:21:08.247885315Z","level":"INFO","msg":"Starting system monitor"}
+{"time":"2025-12-23T19:02:53.999812026Z","level":"INFO","msg":"api: retrying HTTP error","status":502,"url":"https://api.wandb.ai/files/chyang25-national-taiwan-university/box-close-v2-training-split-ablation/4uxuujzc/file_stream"}
+{"time":"2025-12-24T06:32:07.09766907Z","level":"INFO","msg":"stream: closing","id":"4uxuujzc"}
+{"time":"2025-12-24T06:32:07.097695632Z","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2025-12-24T06:32:07.098134463Z","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2025-12-24T06:32:08.120269989Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-24T06:32:08.438751006Z","level":"INFO","msg":"handler: closed","stream_id":{"value":"4uxuujzc"}}
+{"time":"2025-12-24T06:32:08.438784379Z","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"4uxuujzc"}}
+{"time":"2025-12-24T06:32:08.438818549Z","level":"INFO","msg":"sender: closed","stream_id":"4uxuujzc"}
+{"time":"2025-12-24T06:32:08.438851139Z","level":"INFO","msg":"stream: closed","id":"4uxuujzc"}

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug.log

@@ -0,0 +1,27 @@
+2025-12-23 03:21:07,487 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Configure stats pid to 63204
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Loading settings from /root/.config/wandb/settings
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Loading settings from /root/workspace/LLM-BC/wandb/settings
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/root/workspace/LLM-BC/train.py', 'program': '/root/workspace/LLM-BC/./train.py'}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Applying login settings: {}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:_log_setup():533] Logging user logs to /root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/logs/debug.log
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:_log_setup():534] Logging internal logs to /root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/logs/debug-internal.log
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:init():619] calling init triggers
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:init():626] wandb.init called with sweep_config: {}
+config: {'name': 'train_llm_diffusion_unet_lowdim', '_target_': 'llmbc.workspace.train_llm_diffusion_unet_lowdim_workspace.TrainLLMDiffusionUnetLowdimWorkspace', 'obs_dim': 9, 'action_dim': 4, 'task_name': 'box-close-v2', 'exp_name': 'default', 'model_name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'n_latency_steps': 0, 'past_action_visible': False, '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': 4, 'local_cond_dim': None, 'global_cond_dim': 9, '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': 1, 'obs_dim': 9, 'action_dim': 4, 'n_action_steps': 1, 'n_obs_steps': 1, 'num_inference_steps': 100, 'obs_as_local_cond': False, 'obs_as_global_cond': True, 'pred_action_steps_only': False, 'oa_step_convention': True, 'llm_discriminator': {'_target_': 'llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator', 'task_id': 'box-close-v2', 'llm_translator': {'_target_': 'llmbc.translator.llm_translator.LLMTranslator', 'cfg': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.23/03.20.56_HuggingFaceTB/SmolLM2-135M-Instruct'}}}, 'obs_dim': 9, 'action_dim': 4, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1}}, 'loss_dp_weight': 1.0, 'loss_llm_weight': 0.01, 'normalize_llm_loss': 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': 1e-08, 'weight_decay': 1e-06}, 'training': {'device': 'cuda:0', 'seed': 42, 'debug': False, 'resume': True, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 500, 'num_epochs': 20001, 'gradient_accumulate_every': 16, 'use_ema': True, 'rollout_every': 100, 'checkpoint_every': 100, 'val_every': 1, 'sample_every': 100, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-split-ablation', 'resume': True, 'mode': 'online', 'name': '2025.12.23-03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2', 'tags': ['train_llm_diffusion_unet_lowdim', 'box-close-v2', 'default'], 'id': None, 'group': None}, 'checkpoint': {'topk': {'monitor_key': 'test_success_rate', 'mode': 'max', 'k': 5, 'format_str': 'epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt'}, 'save_last_ckpt': True, 'save_last_snapshot': False}, 'multi_run': {'run_dir': 'data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2', 'wandb_name_base': '2025.12.23-03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2'}, 'task': {'name': 'box-close-v2', 'obs_dim': 9, 'action_dim': 4, 'env_runner': {'_target_': 'llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner', 'env_name': 'llf-metaworld-box-close-v2', 'n_train': 10, 'n_test': 50, 'n_envs': 10, 'max_steps': 30, 'n_obs_steps': 1, 'n_action_steps': 1, 'instruction_type': 'b', 'feedback_type': ['hp', 'hn', 'fp'], 'visual': False, 'discount': 0.9}, 'dataset': {'_target_': 'llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset', 'data_path': 'datasets/box-close-v2-split0.05.pt', 'data_path2': 'datasets/box-close-v2-split0.05.pt', 'horizon': 1, 'pad_before': 0, 'pad_after': 0, 'obs_eef_target': True, 'use_manual_normalizer': False, 'val_ratio': 0.1, 'dummy_normalizer': False}, 'instructor': {'_target_': 'llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor'}}, 'llm': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.23/03.20.56_HuggingFaceTB/SmolLM2-135M-Instruct'}}}}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:init():669] starting backend
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:init():673] sending inform_init request
+2025-12-23 03:21:07,489 INFO    MainThread:63204 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2025-12-23 03:21:07,489 INFO    MainThread:63204 [wandb_init.py:init():686] backend started and connected
+2025-12-23 03:21:07,497 INFO    MainThread:63204 [wandb_init.py:init():781] updated telemetry
+2025-12-23 03:21:07,545 INFO    MainThread:63204 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2025-12-23 03:21:08,245 INFO    MainThread:63204 [wandb_init.py:init():867] starting run threads in backend
+2025-12-23 03:21:08,595 INFO    MainThread:63204 [wandb_run.py:_console_start():2451] atexit reg
+2025-12-23 03:21:08,596 INFO    MainThread:63204 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2025-12-23 03:21:08,596 INFO    MainThread:63204 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2025-12-23 03:21:08,596 INFO    MainThread:63204 [wandb_run.py:_redirect():2389] Redirects installed.
+2025-12-23 03:21:08,597 INFO    MainThread:63204 [wandb_init.py:init():911] run started, returning control to user process
+2025-12-23 03:21:08,598 INFO    MainThread:63204 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2'}
+2025-12-24 06:32:07,097 WARNING MsgRouterThr:63204 [router.py:message_loop():75] message_loop has been closed

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/files/config.yaml

@@ -0,0 +1,304 @@
+_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: 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
+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/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+        config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+            run:
+                dir: data/outputs/2025.12.23/03.20.56_HuggingFaceTB/SmolLM2-135M-Instruct
+        llm_mode: pretrained
+        lora_config:
+            bias: none
+            lora_alpha: 64
+            lora_dropout: 0.05
+            r: 32
+            task_type: CAUSAL_LM
+        max_length: 100
+        model_name: SmolLM2-135M-Instruct
+        name: HuggingFaceTB/SmolLM2-135M-Instruct
+        prompter:
+            _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+            use_joint_mlp_projector: true
+        use_joint_mlp_projector: true
+        use_quantization: false
+llm_do_sample:
+    value: false
+llm_orig_expert_feedback:
+    value: true
+logging:
+    value:
+        group: null
+        id: null
+        mode: online
+        name: 2025.12.23-03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2
+        project: box-close-v2-training-split-ablation
+        resume: true
+        tags:
+            - train_llm_diffusion_unet_lowdim
+            - box-close-v2
+            - default
+model_name:
+    value: HuggingFaceTB/SmolLM2-135M-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2
+        wandb_name_base: 2025.12.23-03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2
+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: 9
+optimizer:
+    value:
+        _target_: torch.optim.AdamW
+        betas:
+            - 0.95
+            - 0.999
+        eps: 1e-08
+        lr: 0.0001
+        weight_decay: 1e-06
+output_dir:
+    value: /root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2
+past_action_visible:
+    value: false
+policy:
+    value:
+        _target_: llmbc.policy.llm_diffusion_unet_lowdim_policy.LLMDiffusionUnetLowdimPolicy
+        action_dim: 4
+        horizon: 1
+        llm_discriminator:
+            _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+            llm_translator:
+                _target_: llmbc.translator.llm_translator.LLMTranslator
+                action_dim: 4
+                cfg:
+                    causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+                    checkpoint: data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+                    config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+                        run:
+                            dir: data/outputs/2025.12.23/03.20.56_HuggingFaceTB/SmolLM2-135M-Instruct
+                    llm_mode: pretrained
+                    lora_config:
+                        bias: none
+                        lora_alpha: 64
+                        lora_dropout: 0.05
+                        r: 32
+                        task_type: CAUSAL_LM
+                    max_length: 100
+                    model_name: SmolLM2-135M-Instruct
+                    name: HuggingFaceTB/SmolLM2-135M-Instruct
+                    prompter:
+                        _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+                        use_joint_mlp_projector: true
+                    use_joint_mlp_projector: true
+                    use_quantization: false
+                horizon: 1
+                n_action_steps: 1
+                n_obs_steps: 1
+                obs_dim: 9
+            task_id: box-close-v2
+        loss_dp_weight: 1
+        loss_llm_weight: 0.01
+        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: 9
+            input_dim: 4
+            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
+        normalize_llm_loss: true
+        num_inference_steps: 100
+        oa_step_convention: true
+        obs_as_global_cond: true
+        obs_as_local_cond: false
+        obs_dim: 9
+        pred_action_steps_only: false
+        reweight_llm_loss: true
+pred_action_steps_only:
+    value: false
+task:
+    value:
+        action_dim: 4
+        dataset:
+            _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+            data_path: datasets/box-close-v2-split0.05.pt
+            data_path2: datasets/box-close-v2-split0.05.pt
+            dummy_normalizer: 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.metaworld_lowdim_runner.MetaworldLowdimRunner
+            discount: 0.9
+            env_name: llf-metaworld-box-close-v2
+            feedback_type:
+                - hp
+                - hn
+                - fp
+            instruction_type: b
+            max_steps: 30
+            n_action_steps: 1
+            n_envs: 10
+            n_obs_steps: 1
+            n_test: 50
+            n_train: 10
+            visual: false
+        instructor:
+            _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+        name: box-close-v2
+        obs_dim: 9
+task_name:
+    value: box-close-v2
+training:
+    value:
+        checkpoint_every: 100
+        debug: false
+        device: cuda:0
+        gradient_accumulate_every: 16
+        lr_scheduler: cosine
+        lr_warmup_steps: 500
+        max_train_steps: null
+        max_val_steps: null
+        num_epochs: 20001
+        resume: true
+        rollout_every: 100
+        sample_every: 100
+        seed: 42
+        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

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/files/output.log

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

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/files/wandb-metadata.json

@@ -0,0 +1,56 @@
+{
+  "os": "Linux-4.18.0-513.24.1.el8_9.x86_64-x86_64-with-glibc2.31",
+  "python": "3.9.25",
+  "startedAt": "2025-12-23T03:21:07.489661Z",
+  "args": [
+    "--config-path",
+    "./config/data_efficiency",
+    "--config-name",
+    "llmdp_box-close-v2_pretrained_split0.05.yaml",
+    "policy.loss_llm_weight=1.0e-2"
+  ],
+  "program": "/root/workspace/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "71846e55e50ed33dba16ceb51d4c39784d993ac5"
+  },
+  "email": "chris920325@gmail.com",
+  "root": "/root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2",
+  "host": "llmbc-cuda-545d8cd8bc-9ltjf",
+  "username": "root",
+  "executable": "/root/miniconda3/envs/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 64,
+  "cpu_count_logical": 128,
+  "gpu": "NVIDIA H100 80GB HBM3",
+  "gpu_count": 2,
+  "disk": {
+    "/": {
+      "total": "23041585184768",
+      "used": "586005725184"
+    }
+  },
+  "memory": {
+    "total": "2163619737600"
+  },
+  "cpu": {
+    "count": 64,
+    "countLogical": 128
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    },
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    }
+  ],
+  "cudaVersion": "12.8"
+}

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/files/wandb-summary.json

@@ -0,0 +1 @@
+{"train_loss_dp":0.00013378898438531905,"test/success_rate":0.68,"train/mean_score":8.879635139193185,"test/mean_score":6.43298206980482,"train_loss":0.0001552154717501253,"train/cumulative_reward":16.57029679611994,"global_step":260012,"epoch":20000,"_step":260012,"test/cumulative_reward":12.807193358223481,"_wandb":{"runtime":97859},"lr":9.946691248430285e-13,"_timestamp":1.766557925407844e+09,"val_loss":0.03853258863091469,"_runtime":97859.608022902,"train/success_rate":1,"train_loss_llm":0.0065765380859375,"train_action_mse_error":1.717939213108366e-08}

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2025-12-23T03:21:06.864046036Z","level":"INFO","msg":"started logging, with flags","port-filename":"/tmp/tmp4323r74x/port-63204.txt","pid":63204,"debug":false,"disable-analytics":false}
+{"time":"2025-12-23T03:21:06.864071803Z","level":"INFO","msg":"FeatureState","shutdownOnParentExitEnabled":false}
+{"time":"2025-12-23T03:21:06.881885307Z","level":"INFO","msg":"Will exit if parent process dies.","ppid":63204}
+{"time":"2025-12-23T03:21:06.881889136Z","level":"INFO","msg":"server is running","addr":{"IP":"127.0.0.1","Port":38705,"Zone":""}}
+{"time":"2025-12-23T03:21:07.058428883Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"127.0.0.1:51804"}
+{"time":"2025-12-23T03:21:07.490923354Z","level":"INFO","msg":"handleInformInit: received","streamId":"4uxuujzc","id":"127.0.0.1:51804"}
+{"time":"2025-12-23T03:21:07.600518816Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"4uxuujzc","id":"127.0.0.1:51804"}
+{"time":"2025-12-24T06:32:07.097598027Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"127.0.0.1:51804"}
+{"time":"2025-12-24T06:32:07.097682086Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2025-12-24T06:32:07.097671269Z","level":"INFO","msg":"connection: Close: initiating connection closure","id":"127.0.0.1:51804"}
+{"time":"2025-12-24T06:32:07.097751105Z","level":"INFO","msg":"connection: Close: connection successfully closed","id":"127.0.0.1:51804"}
+{"time":"2025-12-24T06:32:08.438887917Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"127.0.0.1:51804"}
+{"time":"2025-12-24T06:32:08.438903347Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"127.0.0.1:51804"}
+{"time":"2025-12-24T06:32:08.438910872Z","level":"INFO","msg":"server is closed"}

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/logs/debug-internal.log

@@ -0,0 +1,17 @@
+{"time":"2025-12-23T03:21:07.491073659Z","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2025-12-23T03:21:07.491085059Z","level":"INFO","msg":"created symlink","path":"/root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/logs/debug-core.log"}
+{"time":"2025-12-23T03:21:07.600486464Z","level":"INFO","msg":"created new stream","id":"4uxuujzc"}
+{"time":"2025-12-23T03:21:07.600515463Z","level":"INFO","msg":"stream: started","id":"4uxuujzc"}
+{"time":"2025-12-23T03:21:07.600540399Z","level":"INFO","msg":"sender: started","stream_id":"4uxuujzc"}
+{"time":"2025-12-23T03:21:07.600545933Z","level":"INFO","msg":"handler: started","stream_id":{"value":"4uxuujzc"}}
+{"time":"2025-12-23T03:21:07.600531776Z","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"4uxuujzc"}}
+{"time":"2025-12-23T03:21:08.247885315Z","level":"INFO","msg":"Starting system monitor"}
+{"time":"2025-12-23T19:02:53.999812026Z","level":"INFO","msg":"api: retrying HTTP error","status":502,"url":"https://api.wandb.ai/files/chyang25-national-taiwan-university/box-close-v2-training-split-ablation/4uxuujzc/file_stream"}
+{"time":"2025-12-24T06:32:07.09766907Z","level":"INFO","msg":"stream: closing","id":"4uxuujzc"}
+{"time":"2025-12-24T06:32:07.097695632Z","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2025-12-24T06:32:07.098134463Z","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2025-12-24T06:32:08.120269989Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-24T06:32:08.438751006Z","level":"INFO","msg":"handler: closed","stream_id":{"value":"4uxuujzc"}}
+{"time":"2025-12-24T06:32:08.438784379Z","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"4uxuujzc"}}
+{"time":"2025-12-24T06:32:08.438818549Z","level":"INFO","msg":"sender: closed","stream_id":"4uxuujzc"}
+{"time":"2025-12-24T06:32:08.438851139Z","level":"INFO","msg":"stream: closed","id":"4uxuujzc"}

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/logs/debug.log

@@ -0,0 +1,27 @@
+2025-12-23 03:21:07,487 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Configure stats pid to 63204
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Loading settings from /root/.config/wandb/settings
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Loading settings from /root/workspace/LLM-BC/wandb/settings
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/root/workspace/LLM-BC/train.py', 'program': '/root/workspace/LLM-BC/./train.py'}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_setup.py:_flush():79] Applying login settings: {}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:_log_setup():533] Logging user logs to /root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/logs/debug.log
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:_log_setup():534] Logging internal logs to /root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/logs/debug-internal.log
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:init():619] calling init triggers
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:init():626] wandb.init called with sweep_config: {}
+config: {'name': 'train_llm_diffusion_unet_lowdim', '_target_': 'llmbc.workspace.train_llm_diffusion_unet_lowdim_workspace.TrainLLMDiffusionUnetLowdimWorkspace', 'obs_dim': 9, 'action_dim': 4, 'task_name': 'box-close-v2', 'exp_name': 'default', 'model_name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'n_latency_steps': 0, 'past_action_visible': False, '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': 4, 'local_cond_dim': None, 'global_cond_dim': 9, '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': 1, 'obs_dim': 9, 'action_dim': 4, 'n_action_steps': 1, 'n_obs_steps': 1, 'num_inference_steps': 100, 'obs_as_local_cond': False, 'obs_as_global_cond': True, 'pred_action_steps_only': False, 'oa_step_convention': True, 'llm_discriminator': {'_target_': 'llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator', 'task_id': 'box-close-v2', 'llm_translator': {'_target_': 'llmbc.translator.llm_translator.LLMTranslator', 'cfg': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.23/03.20.56_HuggingFaceTB/SmolLM2-135M-Instruct'}}}, 'obs_dim': 9, 'action_dim': 4, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1}}, 'loss_dp_weight': 1.0, 'loss_llm_weight': 0.01, 'normalize_llm_loss': 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': 1e-08, 'weight_decay': 1e-06}, 'training': {'device': 'cuda:0', 'seed': 42, 'debug': False, 'resume': True, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 500, 'num_epochs': 20001, 'gradient_accumulate_every': 16, 'use_ema': True, 'rollout_every': 100, 'checkpoint_every': 100, 'val_every': 1, 'sample_every': 100, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-split-ablation', 'resume': True, 'mode': 'online', 'name': '2025.12.23-03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2', 'tags': ['train_llm_diffusion_unet_lowdim', 'box-close-v2', 'default'], 'id': None, 'group': None}, 'checkpoint': {'topk': {'monitor_key': 'test_success_rate', 'mode': 'max', 'k': 5, 'format_str': 'epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt'}, 'save_last_ckpt': True, 'save_last_snapshot': False}, 'multi_run': {'run_dir': 'data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2', 'wandb_name_base': '2025.12.23-03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2'}, 'task': {'name': 'box-close-v2', 'obs_dim': 9, 'action_dim': 4, 'env_runner': {'_target_': 'llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner', 'env_name': 'llf-metaworld-box-close-v2', 'n_train': 10, 'n_test': 50, 'n_envs': 10, 'max_steps': 30, 'n_obs_steps': 1, 'n_action_steps': 1, 'instruction_type': 'b', 'feedback_type': ['hp', 'hn', 'fp'], 'visual': False, 'discount': 0.9}, 'dataset': {'_target_': 'llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset', 'data_path': 'datasets/box-close-v2-split0.05.pt', 'data_path2': 'datasets/box-close-v2-split0.05.pt', 'horizon': 1, 'pad_before': 0, 'pad_after': 0, 'obs_eef_target': True, 'use_manual_normalizer': False, 'val_ratio': 0.1, 'dummy_normalizer': False}, 'instructor': {'_target_': 'llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor'}}, 'llm': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.23/03.20.56_HuggingFaceTB/SmolLM2-135M-Instruct'}}}}
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:init():669] starting backend
+2025-12-23 03:21:07,488 INFO    MainThread:63204 [wandb_init.py:init():673] sending inform_init request
+2025-12-23 03:21:07,489 INFO    MainThread:63204 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2025-12-23 03:21:07,489 INFO    MainThread:63204 [wandb_init.py:init():686] backend started and connected
+2025-12-23 03:21:07,497 INFO    MainThread:63204 [wandb_init.py:init():781] updated telemetry
+2025-12-23 03:21:07,545 INFO    MainThread:63204 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2025-12-23 03:21:08,245 INFO    MainThread:63204 [wandb_init.py:init():867] starting run threads in backend
+2025-12-23 03:21:08,595 INFO    MainThread:63204 [wandb_run.py:_console_start():2451] atexit reg
+2025-12-23 03:21:08,596 INFO    MainThread:63204 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2025-12-23 03:21:08,596 INFO    MainThread:63204 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2025-12-23 03:21:08,596 INFO    MainThread:63204 [wandb_run.py:_redirect():2389] Redirects installed.
+2025-12-23 03:21:08,597 INFO    MainThread:63204 [wandb_init.py:init():911] run started, returning control to user process
+2025-12-23 03:21:08,598 INFO    MainThread:63204 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/root/workspace/LLM-BC/data/outputs/2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2'}
+2025-12-24 06:32:07,097 WARNING MsgRouterThr:63204 [router.py:message_loop():75] message_loop has been closed

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032107-4uxuujzc/run-4uxuujzc.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d6f771cb37508339f906798177f193be469f0611e586b253a8be9679696d335
+size 278200061

2025.12.23/03.20.56_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "4uxuujzc"}

2025.12.23/03.21.03_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/config.yaml

@@ -0,0 +1,193 @@
+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
+  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: 42
+  debug: false
+  resume: true
+  lr_scheduler: cosine
+  lr_warmup_steps: 500
+  num_epochs: 20001
+  gradient_accumulate_every: 16
+  use_ema: true
+  rollout_every: 100
+  checkpoint_every: 100
+  val_every: 1
+  sample_every: 100
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-split-ablation
+  resume: true
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags:
+  - ${name}
+  - ${task_name}
+  - ${exp_name}
+  id: null
+  group: null
+checkpoint:
+  topk:
+    monitor_key: test_success_rate
+    mode: max
+    k: 5
+    format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+  save_last_ckpt: true
+  save_last_snapshot: false
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+task:
+  name: box-close-v2
+  obs_dim: 9
+  action_dim: 4
+  env_runner:
+    _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+    env_name: llf-metaworld-box-close-v2
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 30
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.9
+  dataset:
+    _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+    data_path: datasets/box-close-v2-split0.05.pt
+    data_path2: datasets/box-close-v2-split0.05.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: false
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: pretrained
+  finetune_mode: orig
+  checkpoint: data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+  max_length: 100
+  lora_config:
+    r: 32
+    lora_alpha: 64
+    lora_dropout: 0.05
+    bias: none
+    task_type: CAUSAL_LM
+  prompter:
+    _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}

2025.12.23/03.21.03_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/hydra.yaml

@@ -0,0 +1,155 @@
+hydra:
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - policy.loss_llm_weight=1.0e-3
+  job:
+    name: train
+    chdir: null
+    override_dirname: policy.loss_llm_weight=1.0e-3
+    id: ???
+    num: ???
+    config_name: llmdp_box-close-v2_pretrained_split0.05.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /root/workspace/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /root/workspace/LLM-BC/config/data_efficiency
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.03_train_llm_diffusion_unet_lowdim_box-close-v2
+    choices:
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

2025.12.23/03.21.03_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/overrides.yaml

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

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/config.yaml

@@ -0,0 +1,193 @@
+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: 0.0001
+  normalize_llm_loss: 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: 42
+  debug: false
+  resume: true
+  lr_scheduler: cosine
+  lr_warmup_steps: 500
+  num_epochs: 20001
+  gradient_accumulate_every: 16
+  use_ema: true
+  rollout_every: 100
+  checkpoint_every: 100
+  val_every: 1
+  sample_every: 100
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-split-ablation
+  resume: true
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags:
+  - ${name}
+  - ${task_name}
+  - ${exp_name}
+  id: null
+  group: null
+checkpoint:
+  topk:
+    monitor_key: test_success_rate
+    mode: max
+    k: 5
+    format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+  save_last_ckpt: true
+  save_last_snapshot: false
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+task:
+  name: box-close-v2
+  obs_dim: 9
+  action_dim: 4
+  env_runner:
+    _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+    env_name: llf-metaworld-box-close-v2
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 30
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.9
+  dataset:
+    _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+    data_path: datasets/box-close-v2-split0.05.pt
+    data_path2: datasets/box-close-v2-split0.05.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: false
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: pretrained
+  finetune_mode: orig
+  checkpoint: data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+  max_length: 100
+  lora_config:
+    r: 32
+    lora_alpha: 64
+    lora_dropout: 0.05
+    bias: none
+    task_type: CAUSAL_LM
+  prompter:
+    _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/hydra.yaml

@@ -0,0 +1,155 @@
+hydra:
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - policy.loss_llm_weight=1.0e-4
+  job:
+    name: train
+    chdir: null
+    override_dirname: policy.loss_llm_weight=1.0e-4
+    id: ???
+    num: ???
+    config_name: llmdp_box-close-v2_pretrained_split0.05.yaml
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.2.0
+    version_base: '1.2'
+    cwd: /root/workspace/LLM-BC
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /root/workspace/LLM-BC/config/data_efficiency
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2
+    choices:
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- policy.loss_llm_weight=1.0e-4

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=2500-test_success_rate=0.800.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd443560289d6900acd9568718324312ff8295fd5c8b440eb3d428e49e666943
+size 1040499058

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=3500-test_success_rate=0.800.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84b2c8d8109fdca1de79d8008e09839c328e92f55b6c71ff2f5abbf0859124e8
+size 1040499058

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=3600-test_success_rate=0.820.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de2461881cd613b58a6fe1f12a0be69e869b8e4f9705357af0396104038a6d06
+size 1040499058

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=4000-test_success_rate=0.800.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d628a8d94917233a1fc9971acadf9877a50d0136c68e7cad1be3b0f12d8e0f13
+size 1040499058

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=4600-test_success_rate=0.820.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9664b94f3e12b1ffc43f1b62fc46e5d529528d36ff45879102c763f3dc1f5a52
+size 1040499058

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb991898b841fb378c53a7b2dd405ee600556fb3116f69abc53ba184b8f136e7
+size 1040499058

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdf09f82b8610cdce545d1dc001ea135512f60a67a7cc7bb0409207075e6ebff
+size 46165495

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/train.log

@@ -0,0 +1,9 @@
+[2025-12-23 03:21:09,167][numexpr.utils][INFO] - Note: detected 128 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable.
+[2025-12-23 03:21:09,167][numexpr.utils][INFO] - Note: NumExpr detected 128 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16.
+[2025-12-23 03:21:09,167][numexpr.utils][INFO] - NumExpr defaulting to 16 threads.
+[2025-12-23 03:21:11,062][datasets][INFO] - PyTorch version 2.2.2 available.
+[2025-12-23 03:21:11,063][datasets][INFO] - TensorFlow version 2.15.1 available.
+[2025-12-23 03:21:11,064][datasets][INFO] - JAX version 0.4.30 available.
+[2025-12-23 03:21:12,551][llmbc.model.diffusion.conditional_unet1d][INFO] - number of parameters: 6.501479e+07
+[2025-12-23 03:21:19,029][absl][INFO] - MUJOCO_GL=osmesa, attempting to import specified OpenGL backend.
+[2025-12-23 03:21:19,034][absl][INFO] - MuJoCo library version is: 2.3.7

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2025-12-23T03:21:19.875607549Z","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2025-12-23T03:21:19.875621659Z","level":"INFO","msg":"created symlink","path":"/root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/logs/debug-core.log"}
+{"time":"2025-12-23T03:21:20.08404388Z","level":"INFO","msg":"created new stream","id":"653bfiyg"}
+{"time":"2025-12-23T03:21:20.084074239Z","level":"INFO","msg":"stream: started","id":"653bfiyg"}
+{"time":"2025-12-23T03:21:20.084095229Z","level":"INFO","msg":"sender: started","stream_id":"653bfiyg"}
+{"time":"2025-12-23T03:21:20.084100606Z","level":"INFO","msg":"handler: started","stream_id":{"value":"653bfiyg"}}
+{"time":"2025-12-23T03:21:20.084094109Z","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"653bfiyg"}}
+{"time":"2025-12-23T03:21:20.838041119Z","level":"INFO","msg":"Starting system monitor"}
+{"time":"2025-12-24T06:28:17.11457334Z","level":"INFO","msg":"stream: closing","id":"653bfiyg"}
+{"time":"2025-12-24T06:28:17.114615011Z","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2025-12-24T06:28:17.115209375Z","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2025-12-24T06:28:18.12268265Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-24T06:28:18.487859989Z","level":"INFO","msg":"handler: closed","stream_id":{"value":"653bfiyg"}}
+{"time":"2025-12-24T06:28:18.487893662Z","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"653bfiyg"}}
+{"time":"2025-12-24T06:28:18.48791428Z","level":"INFO","msg":"sender: closed","stream_id":"653bfiyg"}
+{"time":"2025-12-24T06:28:18.487964813Z","level":"INFO","msg":"stream: closed","id":"653bfiyg"}

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug.log

@@ -0,0 +1,27 @@
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Configure stats pid to 63392
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Loading settings from /root/.config/wandb/settings
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Loading settings from /root/workspace/LLM-BC/wandb/settings
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/root/workspace/LLM-BC/train.py', 'program': '/root/workspace/LLM-BC/./train.py'}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Applying login settings: {}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:_log_setup():533] Logging user logs to /root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/logs/debug.log
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:_log_setup():534] Logging internal logs to /root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/logs/debug-internal.log
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:init():619] calling init triggers
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:init():626] wandb.init called with sweep_config: {}
+config: {'name': 'train_llm_diffusion_unet_lowdim', '_target_': 'llmbc.workspace.train_llm_diffusion_unet_lowdim_workspace.TrainLLMDiffusionUnetLowdimWorkspace', 'obs_dim': 9, 'action_dim': 4, 'task_name': 'box-close-v2', 'exp_name': 'default', 'model_name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'n_latency_steps': 0, 'past_action_visible': False, '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': 4, 'local_cond_dim': None, 'global_cond_dim': 9, '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': 1, 'obs_dim': 9, 'action_dim': 4, 'n_action_steps': 1, 'n_obs_steps': 1, 'num_inference_steps': 100, 'obs_as_local_cond': False, 'obs_as_global_cond': True, 'pred_action_steps_only': False, 'oa_step_convention': True, 'llm_discriminator': {'_target_': 'llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator', 'task_id': 'box-close-v2', 'llm_translator': {'_target_': 'llmbc.translator.llm_translator.LLMTranslator', 'cfg': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.23/03.21.07_HuggingFaceTB/SmolLM2-135M-Instruct'}}}, 'obs_dim': 9, 'action_dim': 4, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1}}, 'loss_dp_weight': 1.0, 'loss_llm_weight': 0.0001, 'normalize_llm_loss': 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': 1e-08, 'weight_decay': 1e-06}, 'training': {'device': 'cuda:0', 'seed': 42, 'debug': False, 'resume': True, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 500, 'num_epochs': 20001, 'gradient_accumulate_every': 16, 'use_ema': True, 'rollout_every': 100, 'checkpoint_every': 100, 'val_every': 1, 'sample_every': 100, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-split-ablation', 'resume': True, 'mode': 'online', 'name': '2025.12.23-03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2', 'tags': ['train_llm_diffusion_unet_lowdim', 'box-close-v2', 'default'], 'id': None, 'group': None}, 'checkpoint': {'topk': {'monitor_key': 'test_success_rate', 'mode': 'max', 'k': 5, 'format_str': 'epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt'}, 'save_last_ckpt': True, 'save_last_snapshot': False}, 'multi_run': {'run_dir': 'data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2', 'wandb_name_base': '2025.12.23-03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2'}, 'task': {'name': 'box-close-v2', 'obs_dim': 9, 'action_dim': 4, 'env_runner': {'_target_': 'llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner', 'env_name': 'llf-metaworld-box-close-v2', 'n_train': 10, 'n_test': 50, 'n_envs': 10, 'max_steps': 30, 'n_obs_steps': 1, 'n_action_steps': 1, 'instruction_type': 'b', 'feedback_type': ['hp', 'hn', 'fp'], 'visual': False, 'discount': 0.9}, 'dataset': {'_target_': 'llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset', 'data_path': 'datasets/box-close-v2-split0.05.pt', 'data_path2': 'datasets/box-close-v2-split0.05.pt', 'horizon': 1, 'pad_before': 0, 'pad_after': 0, 'obs_eef_target': True, 'use_manual_normalizer': False, 'val_ratio': 0.1, 'dummy_normalizer': False}, 'instructor': {'_target_': 'llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor'}}, 'llm': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.23/03.21.07_HuggingFaceTB/SmolLM2-135M-Instruct'}}}}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:init():669] starting backend
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:init():673] sending inform_init request
+2025-12-23 03:21:19,873 INFO    MainThread:63392 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2025-12-23 03:21:19,873 INFO    MainThread:63392 [wandb_init.py:init():686] backend started and connected
+2025-12-23 03:21:19,881 INFO    MainThread:63392 [wandb_init.py:init():781] updated telemetry
+2025-12-23 03:21:19,930 INFO    MainThread:63392 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2025-12-23 03:21:20,835 INFO    MainThread:63392 [wandb_init.py:init():867] starting run threads in backend
+2025-12-23 03:21:21,181 INFO    MainThread:63392 [wandb_run.py:_console_start():2451] atexit reg
+2025-12-23 03:21:21,181 INFO    MainThread:63392 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2025-12-23 03:21:21,181 INFO    MainThread:63392 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2025-12-23 03:21:21,181 INFO    MainThread:63392 [wandb_run.py:_redirect():2389] Redirects installed.
+2025-12-23 03:21:21,183 INFO    MainThread:63392 [wandb_init.py:init():911] run started, returning control to user process
+2025-12-23 03:21:21,183 INFO    MainThread:63392 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2'}
+2025-12-24 06:28:17,114 WARNING MsgRouterThr:63392 [router.py:message_loop():75] message_loop has been closed

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/files/config.yaml

@@ -0,0 +1,304 @@
+_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: 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
+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/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+        config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+        finetune_mode: orig
+        hydra:
+            job:
+                override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+            run:
+                dir: data/outputs/2025.12.23/03.21.07_HuggingFaceTB/SmolLM2-135M-Instruct
+        llm_mode: pretrained
+        lora_config:
+            bias: none
+            lora_alpha: 64
+            lora_dropout: 0.05
+            r: 32
+            task_type: CAUSAL_LM
+        max_length: 100
+        model_name: SmolLM2-135M-Instruct
+        name: HuggingFaceTB/SmolLM2-135M-Instruct
+        prompter:
+            _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+            use_joint_mlp_projector: true
+        use_joint_mlp_projector: true
+        use_quantization: false
+llm_do_sample:
+    value: false
+llm_orig_expert_feedback:
+    value: true
+logging:
+    value:
+        group: null
+        id: null
+        mode: online
+        name: 2025.12.23-03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2
+        project: box-close-v2-training-split-ablation
+        resume: true
+        tags:
+            - train_llm_diffusion_unet_lowdim
+            - box-close-v2
+            - default
+model_name:
+    value: HuggingFaceTB/SmolLM2-135M-Instruct
+multi_run:
+    value:
+        run_dir: data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2
+        wandb_name_base: 2025.12.23-03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2
+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: 9
+optimizer:
+    value:
+        _target_: torch.optim.AdamW
+        betas:
+            - 0.95
+            - 0.999
+        eps: 1e-08
+        lr: 0.0001
+        weight_decay: 1e-06
+output_dir:
+    value: /root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2
+past_action_visible:
+    value: false
+policy:
+    value:
+        _target_: llmbc.policy.llm_diffusion_unet_lowdim_policy.LLMDiffusionUnetLowdimPolicy
+        action_dim: 4
+        horizon: 1
+        llm_discriminator:
+            _target_: llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator
+            llm_translator:
+                _target_: llmbc.translator.llm_translator.LLMTranslator
+                action_dim: 4
+                cfg:
+                    causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+                    checkpoint: data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+                    config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+                    finetune_mode: orig
+                    hydra:
+                        job:
+                            override_dirname: HuggingFaceTB/SmolLM2-135M-Instruct
+                        run:
+                            dir: data/outputs/2025.12.23/03.21.07_HuggingFaceTB/SmolLM2-135M-Instruct
+                    llm_mode: pretrained
+                    lora_config:
+                        bias: none
+                        lora_alpha: 64
+                        lora_dropout: 0.05
+                        r: 32
+                        task_type: CAUSAL_LM
+                    max_length: 100
+                    model_name: SmolLM2-135M-Instruct
+                    name: HuggingFaceTB/SmolLM2-135M-Instruct
+                    prompter:
+                        _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+                        use_joint_mlp_projector: true
+                    use_joint_mlp_projector: true
+                    use_quantization: false
+                horizon: 1
+                n_action_steps: 1
+                n_obs_steps: 1
+                obs_dim: 9
+            task_id: box-close-v2
+        loss_dp_weight: 1
+        loss_llm_weight: 0.0001
+        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: 9
+            input_dim: 4
+            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
+        normalize_llm_loss: true
+        num_inference_steps: 100
+        oa_step_convention: true
+        obs_as_global_cond: true
+        obs_as_local_cond: false
+        obs_dim: 9
+        pred_action_steps_only: false
+        reweight_llm_loss: true
+pred_action_steps_only:
+    value: false
+task:
+    value:
+        action_dim: 4
+        dataset:
+            _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+            data_path: datasets/box-close-v2-split0.05.pt
+            data_path2: datasets/box-close-v2-split0.05.pt
+            dummy_normalizer: 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.metaworld_lowdim_runner.MetaworldLowdimRunner
+            discount: 0.9
+            env_name: llf-metaworld-box-close-v2
+            feedback_type:
+                - hp
+                - hn
+                - fp
+            instruction_type: b
+            max_steps: 30
+            n_action_steps: 1
+            n_envs: 10
+            n_obs_steps: 1
+            n_test: 50
+            n_train: 10
+            visual: false
+        instructor:
+            _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+        name: box-close-v2
+        obs_dim: 9
+task_name:
+    value: box-close-v2
+training:
+    value:
+        checkpoint_every: 100
+        debug: false
+        device: cuda:0
+        gradient_accumulate_every: 16
+        lr_scheduler: cosine
+        lr_warmup_steps: 500
+        max_train_steps: null
+        max_val_steps: null
+        num_epochs: 20001
+        resume: true
+        rollout_every: 100
+        sample_every: 100
+        seed: 42
+        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

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/files/output.log

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

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/files/wandb-metadata.json

@@ -0,0 +1,56 @@
+{
+  "os": "Linux-4.18.0-513.24.1.el8_9.x86_64-x86_64-with-glibc2.31",
+  "python": "3.9.25",
+  "startedAt": "2025-12-23T03:21:19.874112Z",
+  "args": [
+    "--config-path",
+    "./config/data_efficiency",
+    "--config-name",
+    "llmdp_box-close-v2_pretrained_split0.05.yaml",
+    "policy.loss_llm_weight=1.0e-4"
+  ],
+  "program": "/root/workspace/LLM-BC/./train.py",
+  "codePath": "train.py",
+  "git": {
+    "remote": "https://github.com/CHYang25/LLM-BC.git",
+    "commit": "71846e55e50ed33dba16ceb51d4c39784d993ac5"
+  },
+  "email": "chris920325@gmail.com",
+  "root": "/root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2",
+  "host": "llmbc-cuda-545d8cd8bc-9ltjf",
+  "username": "root",
+  "executable": "/root/miniconda3/envs/llm-bc/bin/python3",
+  "codePathLocal": "train.py",
+  "cpu_count": 64,
+  "cpu_count_logical": 128,
+  "gpu": "NVIDIA H100 80GB HBM3",
+  "gpu_count": 2,
+  "disk": {
+    "/": {
+      "total": "23041585184768",
+      "used": "586006028288"
+    }
+  },
+  "memory": {
+    "total": "2163619737600"
+  },
+  "cpu": {
+    "count": 64,
+    "countLogical": 128
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    },
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper"
+    }
+  ],
+  "cudaVersion": "12.8"
+}

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/files/wandb-summary.json

@@ -0,0 +1 @@
+{"lr":9.946691248430285e-13,"epoch":20000,"test/mean_score":6.427205153471092,"train/mean_score":8.354521469948589,"_runtime":97617.240478729,"_timestamp":1.7665576951428096e+09,"train_loss_dp":7.209734758362174e-05,"train_action_mse_error":1.6835885574550957e-08,"test/cumulative_reward":13.018166213730824,"val_loss":0.2622390687465668,"train_loss_llm":0,"train/success_rate":0.9,"_step":260012,"train/cumulative_reward":15.859847059714753,"_wandb":{"runtime":97617},"global_step":260012,"test/success_rate":0.68,"train_loss":0.00016718265340806774}

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2025-12-23T03:21:19.295198403Z","level":"INFO","msg":"started logging, with flags","port-filename":"/tmp/tmpe6vxxomy/port-63392.txt","pid":63392,"debug":false,"disable-analytics":false}
+{"time":"2025-12-23T03:21:19.295221717Z","level":"INFO","msg":"FeatureState","shutdownOnParentExitEnabled":false}
+{"time":"2025-12-23T03:21:19.296703277Z","level":"INFO","msg":"Will exit if parent process dies.","ppid":63392}
+{"time":"2025-12-23T03:21:19.296703368Z","level":"INFO","msg":"server is running","addr":{"IP":"127.0.0.1","Port":45679,"Zone":""}}
+{"time":"2025-12-23T03:21:19.489909956Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"127.0.0.1:40244"}
+{"time":"2025-12-23T03:21:19.875457916Z","level":"INFO","msg":"handleInformInit: received","streamId":"653bfiyg","id":"127.0.0.1:40244"}
+{"time":"2025-12-23T03:21:20.084077738Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"653bfiyg","id":"127.0.0.1:40244"}
+{"time":"2025-12-24T06:28:17.114498414Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"127.0.0.1:40244"}
+{"time":"2025-12-24T06:28:17.114593041Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2025-12-24T06:28:17.114584804Z","level":"INFO","msg":"connection: Close: initiating connection closure","id":"127.0.0.1:40244"}
+{"time":"2025-12-24T06:28:17.114674152Z","level":"INFO","msg":"connection: Close: connection successfully closed","id":"127.0.0.1:40244"}
+{"time":"2025-12-24T06:28:18.488004201Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"127.0.0.1:40244"}
+{"time":"2025-12-24T06:28:18.488018457Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"127.0.0.1:40244"}
+{"time":"2025-12-24T06:28:18.488026407Z","level":"INFO","msg":"server is closed"}

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/logs/debug-internal.log

@@ -0,0 +1,16 @@
+{"time":"2025-12-23T03:21:19.875607549Z","level":"INFO","msg":"using version","core version":"0.18.6"}
+{"time":"2025-12-23T03:21:19.875621659Z","level":"INFO","msg":"created symlink","path":"/root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/logs/debug-core.log"}
+{"time":"2025-12-23T03:21:20.08404388Z","level":"INFO","msg":"created new stream","id":"653bfiyg"}
+{"time":"2025-12-23T03:21:20.084074239Z","level":"INFO","msg":"stream: started","id":"653bfiyg"}
+{"time":"2025-12-23T03:21:20.084095229Z","level":"INFO","msg":"sender: started","stream_id":"653bfiyg"}
+{"time":"2025-12-23T03:21:20.084100606Z","level":"INFO","msg":"handler: started","stream_id":{"value":"653bfiyg"}}
+{"time":"2025-12-23T03:21:20.084094109Z","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"653bfiyg"}}
+{"time":"2025-12-23T03:21:20.838041119Z","level":"INFO","msg":"Starting system monitor"}
+{"time":"2025-12-24T06:28:17.11457334Z","level":"INFO","msg":"stream: closing","id":"653bfiyg"}
+{"time":"2025-12-24T06:28:17.114615011Z","level":"INFO","msg":"Stopping system monitor"}
+{"time":"2025-12-24T06:28:17.115209375Z","level":"INFO","msg":"Stopped system monitor"}
+{"time":"2025-12-24T06:28:18.12268265Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-24T06:28:18.487859989Z","level":"INFO","msg":"handler: closed","stream_id":{"value":"653bfiyg"}}
+{"time":"2025-12-24T06:28:18.487893662Z","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"653bfiyg"}}
+{"time":"2025-12-24T06:28:18.48791428Z","level":"INFO","msg":"sender: closed","stream_id":"653bfiyg"}
+{"time":"2025-12-24T06:28:18.487964813Z","level":"INFO","msg":"stream: closed","id":"653bfiyg"}

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/logs/debug.log

@@ -0,0 +1,27 @@
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Configure stats pid to 63392
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Loading settings from /root/.config/wandb/settings
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Loading settings from /root/workspace/LLM-BC/wandb/settings
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Loading settings from environment variables: {}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Inferring run settings from compute environment: {'program_relpath': 'train.py', 'program_abspath': '/root/workspace/LLM-BC/train.py', 'program': '/root/workspace/LLM-BC/./train.py'}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_setup.py:_flush():79] Applying login settings: {}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:_log_setup():533] Logging user logs to /root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/logs/debug.log
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:_log_setup():534] Logging internal logs to /root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/logs/debug-internal.log
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:init():619] calling init triggers
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:init():626] wandb.init called with sweep_config: {}
+config: {'name': 'train_llm_diffusion_unet_lowdim', '_target_': 'llmbc.workspace.train_llm_diffusion_unet_lowdim_workspace.TrainLLMDiffusionUnetLowdimWorkspace', 'obs_dim': 9, 'action_dim': 4, 'task_name': 'box-close-v2', 'exp_name': 'default', 'model_name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1, 'n_latency_steps': 0, 'past_action_visible': False, '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': 4, 'local_cond_dim': None, 'global_cond_dim': 9, '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': 1, 'obs_dim': 9, 'action_dim': 4, 'n_action_steps': 1, 'n_obs_steps': 1, 'num_inference_steps': 100, 'obs_as_local_cond': False, 'obs_as_global_cond': True, 'pred_action_steps_only': False, 'oa_step_convention': True, 'llm_discriminator': {'_target_': 'llmbc.discriminator.llm_ce_discriminator.LLMCEDiscriminator', 'task_id': 'box-close-v2', 'llm_translator': {'_target_': 'llmbc.translator.llm_translator.LLMTranslator', 'cfg': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.23/03.21.07_HuggingFaceTB/SmolLM2-135M-Instruct'}}}, 'obs_dim': 9, 'action_dim': 4, 'horizon': 1, 'n_obs_steps': 1, 'n_action_steps': 1}}, 'loss_dp_weight': 1.0, 'loss_llm_weight': 0.0001, 'normalize_llm_loss': 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': 1e-08, 'weight_decay': 1e-06}, 'training': {'device': 'cuda:0', 'seed': 42, 'debug': False, 'resume': True, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 500, 'num_epochs': 20001, 'gradient_accumulate_every': 16, 'use_ema': True, 'rollout_every': 100, 'checkpoint_every': 100, 'val_every': 1, 'sample_every': 100, 'max_train_steps': None, 'max_val_steps': None, 'tqdm_interval_sec': 1.0}, 'logging': {'project': 'box-close-v2-training-split-ablation', 'resume': True, 'mode': 'online', 'name': '2025.12.23-03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2', 'tags': ['train_llm_diffusion_unet_lowdim', 'box-close-v2', 'default'], 'id': None, 'group': None}, 'checkpoint': {'topk': {'monitor_key': 'test_success_rate', 'mode': 'max', 'k': 5, 'format_str': 'epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt'}, 'save_last_ckpt': True, 'save_last_snapshot': False}, 'multi_run': {'run_dir': 'data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2', 'wandb_name_base': '2025.12.23-03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2'}, 'task': {'name': 'box-close-v2', 'obs_dim': 9, 'action_dim': 4, 'env_runner': {'_target_': 'llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner', 'env_name': 'llf-metaworld-box-close-v2', 'n_train': 10, 'n_test': 50, 'n_envs': 10, 'max_steps': 30, 'n_obs_steps': 1, 'n_action_steps': 1, 'instruction_type': 'b', 'feedback_type': ['hp', 'hn', 'fp'], 'visual': False, 'discount': 0.9}, 'dataset': {'_target_': 'llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset', 'data_path': 'datasets/box-close-v2-split0.05.pt', 'data_path2': 'datasets/box-close-v2-split0.05.pt', 'horizon': 1, 'pad_before': 0, 'pad_after': 0, 'obs_eef_target': True, 'use_manual_normalizer': False, 'val_ratio': 0.1, 'dummy_normalizer': False}, 'instructor': {'_target_': 'llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor'}}, 'llm': {'name': 'HuggingFaceTB/SmolLM2-135M-Instruct', 'model_name': 'SmolLM2-135M-Instruct', 'config_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig', 'causal_lm_target': 'llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM', 'use_quantization': False, 'use_joint_mlp_projector': True, 'llm_mode': 'pretrained', 'finetune_mode': 'orig', 'checkpoint': 'data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt', 'max_length': 100, 'lora_config': {'r': 32, 'lora_alpha': 64, 'lora_dropout': 0.05, 'bias': 'none', 'task_type': 'CAUSAL_LM'}, 'prompter': {'_target_': 'llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter', 'use_joint_mlp_projector': True}, 'hydra': {'job': {'override_dirname': 'HuggingFaceTB/SmolLM2-135M-Instruct'}, 'run': {'dir': 'data/outputs/2025.12.23/03.21.07_HuggingFaceTB/SmolLM2-135M-Instruct'}}}}
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:init():669] starting backend
+2025-12-23 03:21:19,872 INFO    MainThread:63392 [wandb_init.py:init():673] sending inform_init request
+2025-12-23 03:21:19,873 INFO    MainThread:63392 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2025-12-23 03:21:19,873 INFO    MainThread:63392 [wandb_init.py:init():686] backend started and connected
+2025-12-23 03:21:19,881 INFO    MainThread:63392 [wandb_init.py:init():781] updated telemetry
+2025-12-23 03:21:19,930 INFO    MainThread:63392 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout
+2025-12-23 03:21:20,835 INFO    MainThread:63392 [wandb_init.py:init():867] starting run threads in backend
+2025-12-23 03:21:21,181 INFO    MainThread:63392 [wandb_run.py:_console_start():2451] atexit reg
+2025-12-23 03:21:21,181 INFO    MainThread:63392 [wandb_run.py:_redirect():2299] redirect: wrap_raw
+2025-12-23 03:21:21,181 INFO    MainThread:63392 [wandb_run.py:_redirect():2364] Wrapping output streams.
+2025-12-23 03:21:21,181 INFO    MainThread:63392 [wandb_run.py:_redirect():2389] Redirects installed.
+2025-12-23 03:21:21,183 INFO    MainThread:63392 [wandb_init.py:init():911] run started, returning control to user process
+2025-12-23 03:21:21,183 INFO    MainThread:63392 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/root/workspace/LLM-BC/data/outputs/2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2'}
+2025-12-24 06:28:17,114 WARNING MsgRouterThr:63392 [router.py:message_loop():75] message_loop has been closed

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251223_032119-653bfiyg/run-653bfiyg.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8dbb7b512d96405965995a4190523266094b56ac7f6d43a3d0d5a4411e4f647a
+size 278482691

2025.12.23/03.21.07_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/wandb-resume.json

@@ -0,0 +1 @@
+{"run_id": "653bfiyg"}

2025.12.23/03.21.15_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/config.yaml

@@ -0,0 +1,193 @@
+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
+  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: 42
+  debug: false
+  resume: true
+  lr_scheduler: cosine
+  lr_warmup_steps: 500
+  num_epochs: 20001
+  gradient_accumulate_every: 16
+  use_ema: true
+  rollout_every: 100
+  checkpoint_every: 100
+  val_every: 1
+  sample_every: 100
+  max_train_steps: null
+  max_val_steps: null
+  tqdm_interval_sec: 1.0
+logging:
+  project: ${task.name}-training-split-ablation
+  resume: true
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags:
+  - ${name}
+  - ${task_name}
+  - ${exp_name}
+  id: null
+  group: null
+checkpoint:
+  topk:
+    monitor_key: test_success_rate
+    mode: max
+    k: 5
+    format_str: epoch={epoch:04d}-test_success_rate={test_success_rate:.3f}.ckpt
+  save_last_ckpt: true
+  save_last_snapshot: false
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+task:
+  name: box-close-v2
+  obs_dim: 9
+  action_dim: 4
+  env_runner:
+    _target_: llmbc.env_runner.metaworld_lowdim_runner.MetaworldLowdimRunner
+    env_name: llf-metaworld-box-close-v2
+    n_train: 10
+    n_test: 50
+    n_envs: 10
+    max_steps: 30
+    n_obs_steps: ${n_obs_steps}
+    n_action_steps: ${n_action_steps}
+    instruction_type: b
+    feedback_type:
+    - hp
+    - hn
+    - fp
+    visual: false
+    discount: 0.9
+  dataset:
+    _target_: llmbc.dataset.metaworld_lowdim_dataset.MetaworldLowdimDataset
+    data_path: datasets/box-close-v2-split0.05.pt
+    data_path2: datasets/box-close-v2-split0.05.pt
+    horizon: ${horizon}
+    pad_before: ${eval:'${n_obs_steps}-1'}
+    pad_after: ${eval:'${n_action_steps}-1'}
+    obs_eef_target: true
+    use_manual_normalizer: false
+    val_ratio: 0.1
+    dummy_normalizer: false
+  instructor:
+    _target_: llmbc.translator.instructor.metaworld_instructor.box_close_v2_instructor.BoxCloseV2Instructor
+llm:
+  name: HuggingFaceTB/SmolLM2-135M-Instruct
+  model_name: SmolLM2-135M-Instruct
+  config_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaConfig
+  causal_lm_target: llmbc.model.llm.llama_lowdim_model.LowdimLlamaForCausalLM
+  use_quantization: false
+  use_joint_mlp_projector: true
+  llm_mode: pretrained
+  finetune_mode: orig
+  checkpoint: data/outputs/2025.12.10/12.09.37_train_llm_lowdim_box-close-v2/HuggingFaceTB/SmolLM2-135M-Instruct-finetuned-box-close-v2/normalizer.pt
+  max_length: 100
+  lora_config:
+    r: 32
+    lora_alpha: 64
+    lora_dropout: 0.05
+    bias: none
+    task_type: CAUSAL_LM
+  prompter:
+    _target_: llmbc.translator.prompter.smollm2_prompter.SmolLM2Prompter
+    use_joint_mlp_projector: true
+  hydra:
+    job:
+      override_dirname: ${model_name}
+    run:
+      dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${model_name}