_allow_dotted_keys: value: false _convert_dict: value: true _fields: value: ckpt_steps: 1000 evals: | val: data: name: imagenet2012 split: validation log_steps: 2500 loss_name: softmax_xent pp_fn: decode|resize_small(256)|central_crop(224)|value_range(-1, 1)|onehot(1000, key="label", key_result="labels")|keep("image", "labels") type: classification grad_clip_norm: 1 input: | accum_freq: 8 batch_size: 1024 cache_raw: false data: name: imagenet2012 split: train pp: decode_jpeg_and_inception_crop(224)|flip_lr|randaug(2,10)|value_range(-1, 1)|onehot(1000, key="label", key_result="labels")|keep("image", "labels") shuffle_buffer_size: 150000 log_training_steps: 50 loss: softmax_xent lr: 0.001 mixup: | fold_in: null p: 0.2 model: | pool_type: gap posemb: sincos2d rep_size: false variant: S/16 model_name: vit num_classes: 1000 optax: | mu_dtype: bfloat16 optax_name: scale_by_adam pp_modules: - ops_general - ops_image - ops_text - archive.randaug schedule: | decay_type: cosine warmup_steps: 10000 seed: 0 total_epochs: 90 wd: 0.0001 _locked: value: true _sort_keys: value: true _type_safe: value: true _wandb: value: cli_version: 0.18.7 m: [] python_version: 3.11.10 t: "1": - 1 - 2 - 3 - 12 - 41 - 45 - 55 "2": - 1 - 2 - 3 - 12 - 41 - 45 - 55 "3": - 5 - 13 - 14 - 16 - 23 - 55 - 62 "4": 3.11.10 "5": 0.18.7 "8": - 5 "12": 0.18.7 "13": linux-x86_64