diff --git a/.gitattributes b/.gitattributes index f57b6d8aee483c591c82893ad8825e81084371d2..676af4da59a819b560ded247c4a6d6e796df64f5 100644 --- a/.gitattributes +++ b/.gitattributes @@ -80,3 +80,11 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text 2025.12.24/10.53.06_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251224_105324-0fnjwrs6/run-0fnjwrs6.wandb filter=lfs diff=lfs merge=lfs -text 2025.12.24/10.53.13_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text 2025.12.24/10.53.13_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251224_105328-qrxubtpt/run-qrxubtpt.wandb filter=lfs diff=lfs merge=lfs -text +2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text +2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/run-1t2lmcxa.wandb filter=lfs diff=lfs merge=lfs -text +2025.12.25/14.32.19_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text +2025.12.25/14.32.19_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143231-ftb7uq17/run-ftb7uq17.wandb filter=lfs diff=lfs merge=lfs -text +2025.12.25/14.32.27_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text +2025.12.25/14.32.27_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143243-wud4orbg/run-wud4orbg.wandb filter=lfs diff=lfs merge=lfs -text +2025.12.25/14.32.35_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt filter=lfs diff=lfs merge=lfs -text +2025.12.25/14.32.35_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143251-hkpa7di4/run-hkpa7di4.wandb filter=lfs diff=lfs merge=lfs -text diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/config.yaml b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1107228eb24fbd8728f296fa06c256eaa0d7596e --- /dev/null +++ b/2025.12.25/14.32.12_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: 43 + 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} diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/hydra.yaml b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..204d65cc7974d9531d62b3207aa86df6b5aa5f8d --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/hydra.yaml @@ -0,0 +1,156 @@ +hydra: + run: + dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name} + sweep: + dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - policy.loss_llm_weight=1.0e-3 + - training.seed=43 + job: + name: train + chdir: null + override_dirname: policy.loss_llm_weight=1.0e-3,training.seed=43 + 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/2025.12.25/14.32.12_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 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/overrides.yaml b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f4d268b6714047edf6d4d30b34335ef180cdaf8b --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/.hydra/overrides.yaml @@ -0,0 +1,2 @@ +- policy.loss_llm_weight=1.0e-3 +- training.seed=43 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=2600-test_success_rate=0.820.ckpt b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=2600-test_success_rate=0.820.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4cd0584a151776204101da81fcea67ed2eb6c094 --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=2600-test_success_rate=0.820.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc853fa51f9c610359dca4a1d89d8d52809ec4dc49fa68883ab64b91dd9f820d +size 1040499058 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=3100-test_success_rate=0.820.ckpt b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=3100-test_success_rate=0.820.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..89f7bd61779f8d47e84ccc9db08f60e2fe1f56f2 --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=3100-test_success_rate=0.820.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2c32b060e08610fa088589da8d91049290253d138faf0c1a7d33cd2fc326da3 +size 1040499058 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=4100-test_success_rate=0.820.ckpt b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=4100-test_success_rate=0.820.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..dea7b0fc113d7562eec1cfc477a24148069a6d69 --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=4100-test_success_rate=0.820.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e607598f83b89414a12cf32b59440137795018557158393203ce8cfb38d5abeb +size 1040499058 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=5300-test_success_rate=0.820.ckpt b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=5300-test_success_rate=0.820.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..67771a0838b7af7b8aae1b1f3e58c5db17595a9f --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=5300-test_success_rate=0.820.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0edab21506cd86f435ee992552945bc37fb94a18eb0008aee775606b03bd31cd +size 1040499058 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=5800-test_success_rate=0.820.ckpt b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=5800-test_success_rate=0.820.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..ea00b383170055bf31688f61eb402b2c3397fc74 --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/epoch=5800-test_success_rate=0.820.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da980d90abc83432d54bfcf8580dab6883d081db6a08ab3dd0a733580f032b35 +size 1040499058 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/latest.ckpt b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/checkpoints/latest.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..70dbef88e98565bc55c2d7614defb692a9d50156 --- /dev/null +++ b/2025.12.25/14.32.12_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:f5060a206513e12be784c8a8375a0ceb42e5a5657a0c263664e4e0e17cbdff63 +size 1040499058 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt new file mode 100644 index 0000000000000000000000000000000000000000..88a6dc82bc82698cb3c28642936156813f0dff27 --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/logs.json.txt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89877ef191d3ab2d85c6b6ac914f9f4487772f30e6ac647f4398362f1adefe6d +size 45989648 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/train.log b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/train.log new file mode 100644 index 0000000000000000000000000000000000000000..74247d53312792cf4c23bbef4453f6ae8e8417b8 --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/train.log @@ -0,0 +1,9 @@ +[2025-12-25 14:32:14,266][numexpr.utils][INFO] - Note: detected 128 virtual cores but NumExpr set to maximum of 64, check "NUMEXPR_MAX_THREADS" environment variable. +[2025-12-25 14:32:14,266][numexpr.utils][INFO] - Note: NumExpr detected 128 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 16. +[2025-12-25 14:32:14,266][numexpr.utils][INFO] - NumExpr defaulting to 16 threads. +[2025-12-25 14:32:16,161][datasets][INFO] - PyTorch version 2.2.2 available. +[2025-12-25 14:32:16,162][datasets][INFO] - TensorFlow version 2.15.1 available. +[2025-12-25 14:32:16,163][datasets][INFO] - JAX version 0.4.30 available. +[2025-12-25 14:32:17,725][llmbc.model.diffusion.conditional_unet1d][INFO] - number of parameters: 6.501479e+07 +[2025-12-25 14:32:23,350][absl][INFO] - MUJOCO_GL=osmesa, attempting to import specified OpenGL backend. +[2025-12-25 14:32:23,355][absl][INFO] - MuJoCo library version is: 2.3.7 diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug-internal.log b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug-internal.log new file mode 100644 index 0000000000000000000000000000000000000000..d9172bac234a6427e13c7c5fb3cad5406d9cadcb --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug-internal.log @@ -0,0 +1,16 @@ +{"time":"2025-12-25T14:32:24.206811129Z","level":"INFO","msg":"using version","core version":"0.18.6"} +{"time":"2025-12-25T14:32:24.206821746Z","level":"INFO","msg":"created symlink","path":"/root/workspace/LLM-BC/data/outputs/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/logs/debug-core.log"} +{"time":"2025-12-25T14:32:24.31689039Z","level":"INFO","msg":"created new stream","id":"1t2lmcxa"} +{"time":"2025-12-25T14:32:24.316925876Z","level":"INFO","msg":"stream: started","id":"1t2lmcxa"} +{"time":"2025-12-25T14:32:24.316947718Z","level":"INFO","msg":"sender: started","stream_id":"1t2lmcxa"} +{"time":"2025-12-25T14:32:24.316942108Z","level":"INFO","msg":"writer: Do: started","stream_id":{"value":"1t2lmcxa"}} +{"time":"2025-12-25T14:32:24.31695261Z","level":"INFO","msg":"handler: started","stream_id":{"value":"1t2lmcxa"}} +{"time":"2025-12-25T14:32:24.822195342Z","level":"INFO","msg":"Starting system monitor"} +{"time":"2025-12-26T17:44:43.494520593Z","level":"INFO","msg":"stream: closing","id":"1t2lmcxa"} +{"time":"2025-12-26T17:44:43.49454986Z","level":"INFO","msg":"Stopping system monitor"} +{"time":"2025-12-26T17:44:43.494965654Z","level":"INFO","msg":"Stopped system monitor"} +{"time":"2025-12-26T17:44:44.568834131Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"} +{"time":"2025-12-26T17:44:44.825204851Z","level":"INFO","msg":"handler: closed","stream_id":{"value":"1t2lmcxa"}} +{"time":"2025-12-26T17:44:44.82523139Z","level":"INFO","msg":"sender: closed","stream_id":"1t2lmcxa"} +{"time":"2025-12-26T17:44:44.825228668Z","level":"INFO","msg":"writer: Close: closed","stream_id":{"value":"1t2lmcxa"}} +{"time":"2025-12-26T17:44:44.825316314Z","level":"INFO","msg":"stream: closed","id":"1t2lmcxa"} diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug.log b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug.log new file mode 100644 index 0000000000000000000000000000000000000000..a2eaa4f107c1fe7550ff170c4f9f1949c65d872f --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/debug.log @@ -0,0 +1,27 @@ +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_setup.py:_flush():79] Current SDK version is 0.18.6 +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_setup.py:_flush():79] Configure stats pid to 76737 +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_setup.py:_flush():79] Loading settings from /root/.config/wandb/settings +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_setup.py:_flush():79] Loading settings from /root/workspace/LLM-BC/wandb/settings +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_setup.py:_flush():79] Loading settings from environment variables: {} +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_setup.py:_flush():79] Applying setup settings: {'mode': 'online', '_disable_service': None} +2025-12-25 14:32:24,204 INFO MainThread:76737 [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'} +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_setup.py:_flush():79] Applying login settings: {} +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_init.py:_log_setup():533] Logging user logs to /root/workspace/LLM-BC/data/outputs/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/logs/debug.log +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_init.py:_log_setup():534] Logging internal logs to /root/workspace/LLM-BC/data/outputs/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/logs/debug-internal.log +2025-12-25 14:32:24,204 INFO MainThread:76737 [wandb_init.py:init():619] calling init triggers +2025-12-25 14:32:24,204 INFO MainThread:76737 [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/2025.12.25/14.32.12_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': 0.001, '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': 1e-08, 'weight_decay': 1e-06}, 'training': {'device': 'cuda:0', 'seed': 43, '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': '2025.12.25-14.32.12_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/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2', 'wandb_name_base': '2025.12.25-14.32.12_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/2025.12.25/14.32.12_HuggingFaceTB/SmolLM2-135M-Instruct'}}}} +2025-12-25 14:32:24,205 INFO MainThread:76737 [wandb_init.py:init():669] starting backend +2025-12-25 14:32:24,205 INFO MainThread:76737 [wandb_init.py:init():673] sending inform_init request +2025-12-25 14:32:24,205 INFO MainThread:76737 [backend.py:_multiprocessing_setup():104] multiprocessing start_methods=fork,spawn,forkserver, using: spawn +2025-12-25 14:32:24,206 INFO MainThread:76737 [wandb_init.py:init():686] backend started and connected +2025-12-25 14:32:24,213 INFO MainThread:76737 [wandb_init.py:init():781] updated telemetry +2025-12-25 14:32:24,268 INFO MainThread:76737 [wandb_init.py:init():814] communicating run to backend with 90.0 second timeout +2025-12-25 14:32:24,819 INFO MainThread:76737 [wandb_init.py:init():867] starting run threads in backend +2025-12-25 14:32:25,329 INFO MainThread:76737 [wandb_run.py:_console_start():2451] atexit reg +2025-12-25 14:32:25,329 INFO MainThread:76737 [wandb_run.py:_redirect():2299] redirect: wrap_raw +2025-12-25 14:32:25,329 INFO MainThread:76737 [wandb_run.py:_redirect():2364] Wrapping output streams. +2025-12-25 14:32:25,330 INFO MainThread:76737 [wandb_run.py:_redirect():2389] Redirects installed. +2025-12-25 14:32:25,331 INFO MainThread:76737 [wandb_init.py:init():911] run started, returning control to user process +2025-12-25 14:32:25,332 INFO MainThread:76737 [wandb_run.py:_config_callback():1389] config_cb None None {'output_dir': '/root/workspace/LLM-BC/data/outputs/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2'} +2025-12-26 17:44:43,494 WARNING MsgRouterThr:76737 [router.py:message_loop():75] message_loop has been closed diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/files/config.yaml b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/files/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..21ea7bd088cce4ec0c77718b9b618fae5ae3ecb8 --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/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/2025.12.25/14.32.12_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: 2025.12.25-14.32.12_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/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2 + wandb_name_base: 2025.12.25-14.32.12_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/2025.12.25/14.32.12_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/2025.12.25/14.32.12_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: 0.001 + model: + _target_: llmbc.model.diffusion.conditional_unet1d.ConditionalUnet1D + cond_predict_scale: true + diffusion_step_embed_dim: 256 + down_dims: + - 256 + - 512 + - 1024 + global_cond_dim: 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: 43 + tqdm_interval_sec: 1 + use_ema: true + val_every: 1 +val_dataloader: + value: + batch_size: 16 + num_workers: 0 + persistent_workers: false + pin_memory: false + shuffle: false diff --git a/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/files/output.log b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/files/output.log new file mode 100644 index 0000000000000000000000000000000000000000..1ee9ac8892f73674020284e845e2aab117f863da --- /dev/null +++ b/2025.12.25/14.32.12_train_llm_diffusion_unet_lowdim_box-close-v2/wandb/run-20251225_143224-1t2lmcxa/files/output.log @@ -0,0 +1,3 @@ +Eval MetaworldLowdimRunner 1/6: 0%| | 0/30 [00:00