CONFIG ├── mode │ └── train ├── sudoku │ └── output_dir: /scicore/home/boguno0000/tripod0001/usdm-bench/outputs/sudok │ batch_size: 64 │ accuracy_every_n_steps: 188 │ accuracy_sampling_steps: │ - 8 │ - 16 │ - 32 │ accuracy_num_samples: 256 │ accuracy_batch_size: 64 │ ├── gsm8k │ └── input_file: null │ output_dir: /scicore/home/boguno0000/tripod0001/usdm-bench/outputs/sudok │ timeout: 5.0 │ bootstrap_size: 1 │ ├── seed │ └── 1 ├── allow_tf32 │ └── True ├── cudnn_benchmark │ └── True ├── compile_backbone │ └── False ├── neg_infinity_mode │ └── large-finite ├── loader │ └── global_batch_size: 256 │ eval_global_batch_size: 256 │ batch_size: 256 │ eval_batch_size: 256 │ num_workers: 8 │ pin_memory: true │ ├── confidence_denoiser │ └── confidence_score: entropy │ gamma: 0 │ ├── training │ └── ema: 0.9999 │ antithetic_sampling: true │ importance_sampling: false │ sampling_eps: 0.001 │ change_of_variables: false │ loss_precision: bf16 │ finetune_path: '' │ sampling_eps_min: 0.001 │ sampling_eps_max: 1.0 │ resample: false │ ├── eval │ └── checkpoint_path: '' │ strict_loading: true │ disable_ema: false │ compute_generative_perplexity: false │ perplexity_batch_size: 8 │ compute_perplexity_on_sanity: false │ gen_ppl_eval_model_name_or_path: gpt2-large │ generate_samples: false │ generated_samples_path: /scicore/home/boguno0000/tripod0001/usdm-bench/o │ results_json_path: null │ ├── optim │ └── weight_decay: 0 │ lr: 0.00046938528473548243 │ beta1: 0.9 │ beta2: 0.999 │ eps: 1.0e-08 │ ├── trainer │ └── _target_: lightning.Trainer │ accelerator: cuda │ num_nodes: 1 │ devices: 1 │ accumulate_grad_batches: 1 │ gradient_clip_val: 1.0 │ precision: bf16 │ num_sanity_val_steps: 2 │ max_steps: 20000 │ log_every_n_steps: 10 │ limit_train_batches: 1.0 │ limit_val_batches: 200 │ val_check_interval: 20000 │ check_val_every_n_epoch: null │ ├── wandb │ └── project: debug │ notes: null │ group: null │ job_type: null │ tags: │ - log-linear │ - sudoku │ - sudoku │ - duo_base │ ├── checkpointing │ └── save_dir: /scicore/home/boguno0000/tripod0001/usdm-bench/outputs/sudoku/ │ resume_from_ckpt: true │ resume_ckpt_path: /scicore/home/boguno0000/tripod0001/usdm-bench/outputs │ ├── callbacks │ └── checkpoint_every_n_steps: │ _target_: lightning.pytorch.callbacks.ModelCheckpoint │ save_top_k: -1 │ save_last: true │ dirpath: /scicore/home/boguno0000/tripod0001/usdm-bench/outputs/sudoku │ verbose: true │ auto_insert_metric_name: false │ every_n_train_steps: 5000 │ checkpoint_monitor: │ _target_: lightning.pytorch.callbacks.ModelCheckpoint │ monitor: val/nll │ mode: min │ save_top_k: 1 │ save_last: false │ dirpath: /scicore/home/boguno0000/tripod0001/usdm-bench/outputs/sudoku │ filename: best │ auto_insert_metric_name: false │ verbose: true │ learning_rate_monitor: │ _target_: lightning.pytorch.callbacks.LearningRateMonitor │ logging_interval: step │ ├── data │ └── train: sudoku │ valid: sudoku │ tokenizer_name_or_path: sudoku │ cache_dir: /scicore/home/boguno0000/tripod0001/usdm-bench/data_cache │ wrap: false │ streaming: false │ insert_train_eos: false │ insert_valid_eos: false │ difficulty: easy │ num_train: 48000 │ num_valid: 2000 │ data_seed: 42 │ sudoku_num_workers: 1 │ ├── model │ └── name: tiny │ type: dit │ hidden_size: 512 │ cond_dim: 128 │ length: 180 │ n_blocks: 8 │ n_heads: 8 │ dropout: 0.1 │ ├── strategy │ └── _target_: lightning.pytorch.strategies.DDPStrategy │ find_unused_parameters: false │ ├── noise │ └── type: log-linear │ eps: 0.001 │ alpha_min: null │ alpha_max: null │ adaptive: false │ ├── lr_scheduler │ └── _target_: transformers.get_constant_schedule_with_warmup │ num_warmup_steps: 2500 │ ├── prior │ └── type: none │ latent_width: 0 │ latent_height: 0 │ ├── algo │ └── name: duo_base │ diffusion_type: uniform │ backbone: dit │ parameterization: mean │ time_conditioning: true │ T: 0 │ subs_masking: false │ causal_attention: false │ loss_type: elbo │ adaLN: true │ task_completion: true │ └── sampler └── predictor: ancestral steps: 1024 steps_policy: full noise_removal: ancestral use_float64: true p_nucleus: 1.0 temperature: 1.0 num_sample_batches: 2 num_sample_log: 2