{
  "adaptor_names": null,
  "architectures": [
    "RADIOModel"
  ],
  "args": {
    "aa": null,
    "amp": false,
    "amp_dtype": "float16",
    "amp_impl": "native",
    "aug_repeats": 0,
    "aug_splits": 0,
    "auto_loss_balance_mode": "manual",
    "batch_size": 32,
    "bn_eps": null,
    "bn_momentum": null,
    "cache_dir": null,
    "channels_last": false,
    "checkpoint_hist": 10,
    "chk_keep_forever": 10,
    "class_map": "",
    "clip_grad": null,
    "clip_mode": "norm",
    "cls_token_per_teacher": true,
    "coco_annotations_file": "/datasets/coco2017-adlsa/annotations/captions_val2017.json",
    "coco_image_dir": "/datasets/coco2017-adlsa/val2017",
    "color_jitter": 0.4,
    "cooldown_epochs": 0,
    "cpe_max_size": 2048,
    "crd_loss": false,
    "crd_loss_weight": 0.8,
    "crop_pct": null,
    "cutmix": 0.0,
    "cutmix_minmax": null,
    "data_dir": [
      [
        "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/captioning/datacomp/dc1b/stage2",
        0.95
      ],
      [
        "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/segmentation/sam/stage1",
        0.05
      ]
    ],
    "dataset": "nvgpt4",
    "dataset_download": false,
    "debug_full_knn": false,
    "decay_epochs": 90,
    "decay_milestones": [
      90,
      180,
      270
    ],
    "decay_rate": 0.1,
    "device": "cuda:0",
    "dist_bn": "reduce",
    "distributed": true,
    "drop": 0.0,
    "drop_block": null,
    "drop_connect": null,
    "drop_path": null,
    "dtype": "bfloat16",
    "epoch_repeats": 0.0,
    "epochs": 50,
    "eval": false,
    "eval_metric": "knn_top1",
    "eval_teacher": false,
    "eval_teacher_only": false,
    "eval_throughput": false,
    "experiment": "checkpoints",
    "fast_norm": false,
    "feature_summarizer": "cls_token",
    "feature_upscale_factor": null,
    "force_new_wandb_id": false,
    "force_spectral_reparam": false,
    "fuser": "",
    "gp": null,
    "grad_accum_steps": 1,
    "grad_checkpointing": false,
    "head_init_bias": null,
    "head_init_scale": null,
    "hflip": 0.5,
    "img_size": null,
    "in_chans": null,
    "initial_checkpoint": "",
    "input_size": null,
    "interpolation": "",
    "layer_decay": null,
    "local_rank": 0,
    "log_interval": 50,
    "log_mlflow": false,
    "log_wandb": true,
    "loss": "cosine",
    "loss_auto_balance": false,
    "lr": 0.001,
    "lr_base": 0.1,
    "lr_base_scale": "",
    "lr_base_size": 256,
    "lr_cycle_decay": 0.5,
    "lr_cycle_limit": 1,
    "lr_cycle_mul": 1.0,
    "lr_k_decay": 1.0,
    "lr_noise": null,
    "lr_noise_pct": 0.67,
    "lr_noise_std": 1.0,
    "mean": null,
    "mesa": {
      "gaussian_kl": false,
      "start_epoch": 100
    },
    "min_lr": 0,
    "mixup": 0.0,
    "mixup_mode": "batch",
    "mixup_off_epoch": 0,
    "mixup_prob": 1.0,
    "mixup_switch_prob": 0.5,
    "mlp_hidden_size": 1520,
    "mlp_num_inner": 3,
    "mlp_version": "v2",
    "model": "vit_huge_patch16_224_mlpnorm",
    "model_ema": {
      "decay": 0.9998,
      "force_cpu": false,
      "power": false,
      "power_stds": [
        0.05,
        0.1
      ],
      "start_epoch": 0
    },
    "model_kwargs": {},
    "model_norm": true,
    "momentum": 0.9,
    "no_aug": false,
    "no_ddp_bb": false,
    "no_prefetcher": false,
    "no_resume_opt": false,
    "num_classes": null,
    "opt": "lamb",
    "opt_betas": null,
    "opt_eps": null,
    "opt_kwargs": {
      "filter_bias_and_bn": false
    },
    "output": "/lustre/fs6/portfolios/llmservice/users/mranzinger/output/evfm/ohem/3-13-24_vit-h-16_bf16_ep50",
    "patience_epochs": 10,
    "pin_mem": false,
    "prefetcher": true,
    "pretrained": false,
    "rank": 0,
    "ratio": [
      0.75,
      1.3333333333333333
    ],
    "recount": 1,
    "recovery_interval": 0,
    "register_multiple": 16,
    "remode": "pixel",
    "reprob": 0.0,
    "resplit": false,
    "resume": "/lustre/fs6/portfolios/llmservice/users/mranzinger/output/evfm/ohem/3-13-24_vit-h-16_bf16_ep50/checkpoints/checkpoint-48.pth.tar",
    "save_images": false,
    "scale": [
      0.5,
      1.0
    ],
    "sched": "cosine",
    "sched_on_updates": true,
    "seed": 42,
    "smoothing": 0.1,
    "spectral_reparam": false,
    "split_bn": false,
    "start_epoch": null,
    "std": null,
    "steps_per_epoch": 2000,
    "sync_bn": false,
    "synchronize_step": true,
    "teachers": [
      {
        "amp": true,
        "amp_dtype": "bfloat16",
        "batch_size": 16,
        "data_dir": [
          [
            "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/captioning/datacomp/dc1b/stage2",
            0.95
          ],
          [
            "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/segmentation/sam/stage1",
            0.05
          ]
        ],
        "fd_loss_weight": 1.0,
        "fd_normalize": false,
        "feature_distillation": true,
        "input_size": 378,
        "match_pre_proj": false,
        "model": "ViT-H-14-378-quickgelu",
        "name": "clip",
        "pretrained": "dfn5b",
        "sample_rate": 16,
        "student_resolution": 432,
        "summary_loss_weight": 1.0,
        "torchcompile": true,
        "type": "open_clip",
        "vitdet_prob": 0.05,
        "vitdet_window_sizes": [
          3,
          9,
          9,
          9
        ]
      },
      {
        "amp": false,
        "amp_dtype": "bfloat16",
        "batch_size": 16,
        "fd_loss_weight": 0.8,
        "fd_normalize": false,
        "feature_distillation": true,
        "input_size": 336,
        "match_pre_proj": false,
        "model": "ViT-L/14@336px",
        "name": "openai_clip",
        "pretrained": "openai",
        "sample_rate": 16,
        "summary_loss_weight": 0.8,
        "torchcompile": true,
        "type": "openai_clip",
        "use_summary": false
      },
      {
        "amp": true,
        "amp_dtype": "bfloat16",
        "batch_size": 16,
        "fd_loss_weight": 2.0,
        "fd_normalize": false,
        "feature_distillation": true,
        "input_size": 378,
        "model": "dinov2_vitg14_reg",
        "name": "dino_v2",
        "sample_rate": 16,
        "summary_loss_weight": 1.0,
        "torchcompile": true,
        "type": "dino_v2"
      },
      {
        "amp": false,
        "batch_size": 2,
        "data_dir": [
          [
            "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/segmentation/sam/stage1",
            0.4
          ]
        ],
        "fd_loss_fn": "MSE",
        "fd_loss_weight": 0.25,
        "fd_normalize": false,
        "fd_ohem": true,
        "feature_distillation": true,
        "input_size": 1024,
        "model": "vit-h",
        "name": "sam",
        "sample_rate": 2,
        "student_resolution": 1024,
        "summary_loss_weight": 1e-05,
        "type": "sam",
        "use_summary": false,
        "vitdet_prob": 0.99,
        "vitdet_window_sizes": [
          8,
          16,
          16
        ]
      },
      {
        "amp": true,
        "batch_size": 2,
        "data_dir": [
          [
            "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/ocr/publaynet/webdataset",
            0.4
          ],
          [
            "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/ocr/staging/arxiv/hocr",
            0.4
          ],
          [
            "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/ocr/scene-text/scene-text/text_ocr/webdataset",
            0.15
          ],
          [
            "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/ocr/scene-text/scene-text/hiertext/webdataset",
            0.05
          ]
        ],
        "fd_loss_fn": "MSE",
        "fd_loss_weight": 0.13,
        "fd_normalize": false,
        "fd_ohem": true,
        "fd_upsample_factor": 4,
        "feature_distillation": true,
        "input_size": 1024,
        "model": "quality",
        "name": "rtx-translate",
        "sample_rate": 2,
        "student_resolution": 1024,
        "summary_loss_weight": 1e-05,
        "type": "rtx_translate",
        "use_summary": false,
        "vitdet_prob": 0.99,
        "vitdet_window_sizes": [
          8,
          16,
          16
        ]
      }
    ],
    "torchcompile": null,
    "torchscript": false,
    "train_interpolation": "random",
    "train_split": "train",
    "tta": 0,
    "use_coco": false,
    "use_multi_epochs_loader": false,
    "val_data_dir": "/lustre/fsw/portfolios/llmservice/projects/llmservice_nlp_fm/datasets/classification/imagenet-1k/webdataset",
    "val_ema_only": false,
    "val_img_size": 432,
    "val_jobs_script": "run_validation_jobs_vit-h-16.sh",
    "val_split": "val",
    "validation_batch_size": 64,
    "vflip": 0.0,
    "wandb_entity": "",
    "wandb_group": "ohem",
    "wandb_job_type": "",
    "wandb_name": "",
    "wandb_project": "",
    "warmup_epochs": 0.5,
    "warmup_lr": 1e-05,
    "warmup_prefix": false,
    "weight_decay": 0.02,
    "worker_seeding": "all",
    "workers": 8,
    "world_size": 128
  },
  "auto_map": {
    "AutoConfig": "hf_model.RADIOConfig",
    "AutoModel": "hf_model.RADIOModel"
  },
  "max_resolution": 2048,
  "patch_size": 16,
  "preferred_resolution": [
    432,
    432
  ],
  "torch_dtype": "bfloat16",
  "transformers_version": "4.40.1",
  "version": "radio_v2.1",
  "vitdet_window_size": null
}