Tao Train Dinov3

NVIDIAOfficialCreated by NVIDIA Corporation

Skill
Tool integration
Data and AI
Apache-2.0
Medium risk

From the author

DINOv3 continual self-supervised pre-training. Domain-adapts public DINOv3 ViT backbones on unlabeled images via teacher-student self-distillation (DINO + iBOT + KoLeo, optional Gram anchoring) and converts the EMA teacher into a timm-format backbone for downstream tasks. Trigger phrases include "train DINOv3", "DINOv3 SSL", "domain-adapt a foundation backbone", "continual pretraining", "self-supervised finetune DINOv3".

— the description by NVIDIA Corporation

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

The tao-train-dinov3 skill provides a workflow for domain-adapting DINOv3 vision transformer backbones using self-supervised learning on unlabeled image data. It employs teacher-student self-distillation techniques, including DINO, iBOT, and KoLeo objectives, with optional Gram anchoring. The skill supports full training or parameter-efficient fine-tuning via LoRA, and includes utilities to convert, export, and run inference with trained teacher checkpoints.

What it does

Performs continual self-supervised pre-training on unlabeled image datasets.
Supports teacher-student self-distillation using DINO, iBOT, and KoLeo objectives.
Enables parameter-efficient fine-tuning using LoRA adapters.
Provides tools to convert, export, and perform inference with trained teacher checkpoints.
Supports AutoML for automated hyperparameter optimization.
Allows for high-resolution training configurations.

When to use it

When you need to adapt a pre-trained DINOv3 backbone to a specific domain using unlabeled data.
When you want to improve representation quality for downstream tasks like classification, detection, or segmentation.
When you need to perform continual learning on a foundation model.
When you require a standardized workflow to export or convert DINOv3 models for deployment.

Advantages

Leverages advanced self-supervised objectives for robust feature learning.
Supports LoRA for efficient adaptation without full parameter updates.
Includes built-in support for AutoML to simplify hyperparameter tuning.
Provides clear separation between training, evaluation, and deployment artifacts.
Compatible with standard timm-format backbones.

Usage examples

“Train a DINOv3 backbone on my unlabeled dataset.”: The agent initiates a training run using the specified configuration, applying DINO/iBOT objectives to adapt the backbone to the provided images.
“Export my trained DINOv3 teacher checkpoint for deployment.”: The agent processes the selected teacher checkpoint and converts it into the required format for downstream inference.

Install

Download the skill as a ZIP:
https://mp-staging.autogineer.com/agent-skills/nvidia/tao-train-dinov3/download

Unzip it into the cross-client skills folder of your project:

mkdir -p .agents/skills && unzip tao-train-dinov3.zip -d .agents/skills

Documentation by NVIDIA Corporation

DINOv3

Standalone install? If this session was not initialized by the TAO skill bank plugin, run the tao-setup skill first.

Use this skill to continue self-supervised training of a DINOv3 backbone on unlabeled images, then convert, export, or run inference with a trained teacher checkpoint.

Quick Start (docker run)

Docker-native launch — no TAO SDK and no Python on the host. Use the local Docker/platform skill instead when it gives a stricter environment-specific command (non-root UID mapping, cache redirects, remote daemons).

TAO_PYT_IMAGE_DEFAULT=nvcr.io/nvidia/tao/tao-toolkit:7.3.0-pyt  # versions-key: images.tao_toolkit.pyt
TAO_PYT_IMAGE="${TAO_PYT_IMAGE:-$TAO_PYT_IMAGE_DEFAULT}"
RUN_ROOT="${RUN_ROOT:-$PWD}"
DOCKER_COMMON=(
  --rm --gpus all --shm-size=8g
  --shm-size=8g
  --ulimit memlock=-1
  --ulimit stack=67108864
  -v "$RUN_ROOT/data:/data:ro"
  -v "$RUN_ROOT/specs:/specs:ro"
  -v "$RUN_ROOT/results:/results"
)

Train:

docker run "${DOCKER_COMMON[@]}" "$TAO_PYT_IMAGE" \
  dinov3 train -e /specs/train.yaml

Inference:

docker run "${DOCKER_COMMON[@]}" "$TAO_PYT_IMAGE" \
  dinov3 inference -e /specs/inference.yaml

Export:

docker run "${DOCKER_COMMON[@]}" "$TAO_PYT_IMAGE" \
  dinov3 export -e /specs/export.yaml

Convert:

docker run "${DOCKER_COMMON[@]}" "$TAO_PYT_IMAGE" \
  dinov3 convert -e /specs/convert.yaml

Every action takes its spec with -e; results_dir is set in the spec or overridden on the command line. Mount any pretrained-weights directory the spec references, and keep every in-container path consistent across actions.

Configuration

Use schemas/<action>.schema.json for supported fields, defaults, ranges, and options. Start from the matching references/spec_template_<action>.yaml.

Training is AutoML-enabled. Read references/skill_info.yaml and resolve an explicit automl_policy or user request before training. Default to automl_policy: on; requests such as "disable AutoML", "no HPO", or "plain training" mean off for that run. When it is on and the train schema and template are packaged, route training through tao-skill-bank:tao-run-automl. Use direct training only when the policy is off or those files are missing, and report the missing-file limitation. Non-train actions remain in this skill.

Treat train_loss as an optimization signal; use downstream evaluation when selecting a checkpoint for deployment.

For method background or tuning ideas, optionally read:

references/dinov3-method.md
references/dinov3-recipes.md

Core workflow

Obtain a compatible DINOv3 checkpoint in timm or TAO format.
Point dataset.train_dataset.images_dir to a directory of unlabeled images.
Configure the backbone, training scale, and output location.
Run training and evaluate useful teacher checkpoints on data representative of the downstream task.
Use the selected teacher checkpoint with convert, export, or inference.

Provide either train.pretrained_model_path for a new run or train.resume_training_checkpoint_path to resume an existing run.

Essential fields

Purpose	Spec field
Training images	`dataset.train_dataset.images_dir`
Initial weights	`train.pretrained_model_path`
Resume checkpoint	`train.resume_training_checkpoint_path`
Backbone	`model.backbone.teacher_type`, `model.backbone.student_type`
Training scale	`dataset.batch_size`, `train.num_gpus`, `train.num_nodes`
Output directory	`results_dir`
Conversion input	`convert.checkpoint`
Export input	`export.checkpoint`
Inference inputs	`inference.checkpoint`, `dataset.test_dataset.images_dir`

Example train settings:

model:
  backbone:
    teacher_type: vit_b
    student_type: vit_b
    rope_theta: 100.0
dataset:
  train_dataset:
    images_dir: /path/to/unlabeled/images
  batch_size: 16
train:
  pretrained_model_path: /path/to/dinov3/weights
  num_epochs: 10
  num_gpus: 1

Supported image extensions include jpg, jpeg, png, ppm, bmp, pgm, tif, tiff, and webp. Keep evaluation data separate from training data when measuring downstream quality.

Every dataset.*.images_dir value must be an extracted image folder. Managed archive-backed sources use runtime: extracted_folder, which unpacks the source before injecting the directory into the spec. For direct Docker or TAO CLI runs, extract .tar, .tar.gz, .tgz, or .zip inputs before launch.

Checkpoints

File	Use
`model_epoch__step_.pth`	Resume training, or export its EMA teacher
`teacher_epoch__step_.pth`	Convert, export, inference, or downstream evaluation
`student_epoch__step_.pth`	Diagnostics

For export, prefer a selected stripped teacher_epoch_*_step_*.pth. Export also accepts a selected full model_epoch_*_step_*.pth and deterministically selects and logs teacher.backbone. Inference requires a stripped teacher checkpoint. Do not substitute dinov3_model_latest.pth without evaluating that milestone.

Training loss does not directly measure downstream representation quality. When possible, compare candidate teacher checkpoints using metrics relevant to the target task.

For convert, export, and inference, copy the training-time backbone configuration into the action spec, including rope_theta. Stripped checkpoints do not restore the RoPE frequency base, so falling back to the default can change the trained model's behavior.

Key options

model.backbone.teacher_type and student_type support vit_s, vit_s_plus, vit_b, vit_l, vit_h_plus, and vit_7b.
When model.distill.enable: false, the teacher and student backbone types must match.
model.backbone.rope_theta controls the rotary frequency base. It defaults to 100.0 and is tunable; changing it alters positional encoding, so evaluate the resulting checkpoint for the intended task.
model.centering_method supports sinkhorn and softmax.
model.gram.* controls optional Gram anchoring.
model.lora.* enables parameter-efficient continual pre-training. The default targets the attention qkv and proj projections; num_last_blocks: 0 adapts every transformer block.
model.preservation.* adds CLS-token MSE and cosine losses against the frozen anchor teacher. It is recommended with LoRA when global embedding quality matters.
train.log_every_n_steps controls step-level loss visibility and defaults to 1, so short smoke runs emit component and preservation losses.
train.precision defaults to 16-mixed; choose precision based on hardware and measured quality.
train.use_custom_attention: false uses native attention when custom attention is unavailable.
Crop and export dimensions need to be divisible by model.backbone.patch_size.

Distributed training

For iterative GRIT or multi-task weakness mining from an unlabeled pool followed by DINOv3 SSL retraining, use the related tao-run-dinov3-ssl-deft application. Ordinary pretraining and fine-tuning remain in this model skill.

Use train.num_gpus, train.gpu_ids, train.num_nodes, and train.distributed_strategy. Start with auto; use ddp or fsdp when appropriate for the selected backbone, resolution, and available memory.

Common issues

Checkpoint load failure: verify that the checkpoint is DINOv3 and that its backbone matches the configured teacher and student types.
No images found or archive rejected: pass an extracted image folder to dataset.*.images_dir, not an archive file.
Out of memory: reduce batch size or image resolution, or use a sharded distributed strategy.
Custom-attention failure: set train.use_custom_attention: false.
Deterministic-backward failure: xformers memory-efficient attention does not provide a deterministic backward kernel; use train.cudnn.deterministic: false.
Grid-size error: choose crop and export dimensions divisible by the patch size.
Unexpected inference keys: use a stripped teacher_epoch_*_step_*.pth checkpoint.
Full-checkpoint export: a selected model_epoch_*_step_*.pth is supported; export selects and logs its EMA teacher.backbone.
Resume mismatch: the SSL dataloader is not step-resumable, so resume from an epoch/checkpoint boundary when reproducibility matters.

Files

ATTRIBUTION.mdAdded by AutoGineer918 B
BENCHMARK.md7.1 KiB
LICENSELicence file added by AutoGineer11.2 KiB
SKILL.md8.8 KiB
config/skillspector-baseline.yaml1.0 KiB
evals/evals.json2.2 KiB
references/dinov3-method.md2.3 KiB
references/dinov3-recipes.md2.9 KiB
references/skill_info.yaml3.5 KiB
references/spec_template_convert.yaml2.6 KiB
references/spec_template_export.yaml2.5 KiB
references/spec_template_inference.yaml2.3 KiB
references/spec_template_train.yaml3.9 KiB
references/spec_template_train_highres.yaml3.3 KiB
schemas/convert.schema.json56.0 KiB
schemas/export.schema.json56.9 KiB
schemas/inference.schema.json56.9 KiB
schemas/manifest.json19.8 KiB
schemas/train.schema.json54.1 KiB
skill-card.md4.7 KiB
skill.oms.sig7.8 KiB

schemas/convert.schema.json

{
  "automl_default_parameters": [
    "dataset.workers",
    "dataset.batch_size"
  ],
  "automl_disabled_parameters": [
    "dataset.train_dataset",
    "wandb.tags",
    "inference",
    "train.cudnn",
    "train.gpu_ids",
    "train.schedulers",
    "model",
    "model.head",
    "model.backbone",
    "dataset",
    "train.checkpointer",
    "train.optim",
    "model.gram",
    "gen_trt_engine",
    "model.preservation",
    "train.schedulers.weight_decay",
    "model.lora.target_modules",
    "train.schedulers.last_layer_learning_rate",
    "dataset.transform.global_crops_scale",
    "train.schedulers.learning_rate",
    "wandb",
    "model.distill",
    "train",
    "train.schedulers.teacher_temperature",
    "inference.gpu_ids",
    "dataset.transform",
    "train.schedulers.momentum",
    "gen_trt_engine.tensorrt.layers_precision",
    "dataset.transform.local_crops_scale",
    "gen_trt_engine.tensorrt",
    "convert",
    "dataset.test_dataset",
    "model.lora",
    "export"
  ],
  "default": {
    "convert": {
      "checkpoint": "",
      "output_path": "",
      "results_dir": "",
      "source": "teacher",
      "validate": true
    },
    "dataset": {
      "batch_size": 4,
      "pin_memory": true,
      "test_dataset": {
        "images_dir": ""
      },
      "train_dataset": {
        "images_dir": ""
      },
      "transform": {
        "global_crops_scale": [
          0.32,
          1.0
        ],
        "global_crops_size": 256,
        "local_crops_scale": [
          0.05,
          0.32
        ],
        "local_crops_size": 112,
        "n_global_crops": 2,
        "n_local_crops": 8
      },
      "workers": 8
    },
    "encryption_key": "",
    "model": {
      "backbone": {
        "drop_path_rate": 0.4,
        "img_size": 256,
        "num_register_tokens": 4,
        "patch_size": 16,
        "rope_theta": 100.0,
        "student_type": "vit_b",
        "teacher_type": "vit_b"
      },
      "centering_method": "sinkhorn",
      "distill": {
        "disable_masking": false,
        "enable": false,
        "pretrained_non_distill_pl_model_path": ""
      },
      "gram": {
        "enable": false,
        "refresh_interval": 0,
        "start_step": 0,
        "teacher_scale": 1.0,
        "teacher_source": "pretrained",
        "w_gram": 0.0
      },
      "head": {
        "bottleneck_dim": 384,
        "hidden_dim": 2048,
        "num_layers": 3
      },
      "lora": {
        "alpha": 16.0,
        "dropout": 0.05,
        "enable": false,
        "num_last_blocks": 0,
        "rank": 8,
        "target_modules": [
          "qkv",
          "proj"
        ]
      },
      "preservation": {
        "cls_cosine_weight": 0.05,
        "cls_mse_weight": 0.05,
        "enable": false
      }
    },
    "model_name": "",
    "results_dir": "",
    "train": {
      "checkpoint_interval": 1,
      "checkpoint_interval_unit": "epoch",
      "checkpointer": {
        "auto_insert_metric_name": false,
        "enable_topk": false,
        "filename": "model_best_{epoch:03d}",
        "replace_periodic": false,
        "save_top_k": 1
      },
      "clip_grad_norm": 3.0,
      "cudnn": {
        "benchmark": true,
        "deterministic": false
      },
      "distributed_strategy": "auto",
      "gpu_ids": [
        0
      ],
      "layerwise_decay": 1.0,
      "log_every_n_steps": 1,
      "num_epochs": 10,
      "num_gpus": 1,
      "num_nodes": 1,
      "num_prototypes": 131072,
      "optim": {
        "optim": "adamw"
      },
      "precision": "16-mixed",
      "results_dir": "",
      "resume_training_checkpoint_path": "",
      "schedulers": {
        "last_layer_learning_rate": {
          "freeze_steps": 1250,
          "max_decay_steps": 2500000,
          "val_base": 7.07e-06,
          "val_final": 1e-06,
          "val_start": 0.0,
          "warm_up_steps": 100000
        },
        "learning_rate": {
          "max_decay_steps": 2500000,
          "val_base": 7.07e-06,
          "val_final": 1e-06,
          "val_start": 0.0,
          "warm_up_steps": 100000
        },
        "momentum": {
          "max_decay_steps": 2500000,
          "val_base": 0.994,
          "val_final": 1.0,
          "val_start": 0.0,
          "warm_up_steps": 0
        },
        "teacher_temperature": {
          "max_decay_steps": 37500,
          "val_base": 0.07,
          "val_final": 0.07,
          "val_start": 0.04,
          "warm_up_steps": 37500
        },
        "weight_decay": {
          "max_decay_steps": 2500000,
          "val_base": 0.04,
          "val_final": 0.2,
          "val_start": 0.0,
          "warm_up_steps": 0
        }
      },
      "seed": 1234,
      "use_custom_attention": true,
      "validation_interval": 1
    },
    "wandb": {
      "enable": true,
      "entity": "",
      "group": "",
      "name": "TAO Toolkit Training",
      "project": "TAO Toolkit",
      "reinit": false,
      "run_id": "",
      "save_code": false,
      "sync_tensorboard": false,
      "tags": [
        "tao-toolkit"
      ]
    }
  },
  "popular": {
    "convert": {
      "checkpoint": "",
      "output_path": "",
      "source": "teacher"
    },
    "dataset": {
      "batch_size": 4,
      "pin_memory": true,
      "transform": {
        "global_crops_scale": [
          0.32,
          1.0
        ],
        "global_crops_size": 256,
        "local_crops_scale": [
          0.05,
          0.32
        ],
        "local_crops_size": 112,
        "n_global_crops": 2,
        "n_local_crops": 8
      },
      "workers": 8
    },
    "gen_trt_engine": {
      "batch_size": -1,
      "gpu_id": 0,
      "tensorrt": {
        "max_batch_size": 1,
        "min_batch_size": 1,
        "opt_batch_size": 1
      }
    },
    "inference": {
      "gpu_ids": [
        0
      ],
      "num_gpus": 1,
      "num_nodes": 1
    },
    "model": {
      "backbone": {
        "drop_path_rate": 0.4,
        "img_size": 256,
        "num_register_tokens": 4,
        "patch_size": 16,
        "rope_theta": 100.0
      },
      "centering_method": "sinkhorn",
      "distill": {
        "disable_masking": false,
        "enable": false
      },
      "gram": {
        "enable": false,
        "start_step": 0,
        "w_gram": 0.0
      },
      "head": {
        "bottleneck_dim": 384,
        "hidden_dim": 2048,
        "num_layers": 3
      },
      "lora": {
        "alpha": 16.0,
        "enable": false,
        "num_last_blocks": 0,
        "rank": 8,
        "target_modules": [
          "qkv",
          "proj"
        ]
      },
      "preservation": {
        "cls_cosine_weight": 0.05,
        "cls_mse_weight": 0.05,
        "enable": false
      }
    },
    "train": {
      "checkpoint_interval": 1,
      "clip_grad_norm": 3.0,
      "distributed_strategy": "auto",
      "gpu_ids": [
        0
      ],
      "layerwise_decay": 1.0,
      "num_epochs": 10,
      "num_gpus": 1,
      "num_nodes": 1,
      "num_prototypes": 131072,
      "optim": {
        "optim": "adamw"
      },
      "precision": "16-mixed",
      "schedulers": {
        "last_layer_learning_rate": {
          "freeze_steps": 1250,
          "max_decay_steps": 2500000,
          "val_base": 7.07e-06,
          "val_final": 1e-06,
          "val_start": 0.0,
          "warm_up_steps": 100000
        },
        "learning_rate": {
          "max_decay_steps": 2500000,
          "val_base": 7.07e-06,
          "val_final": 1e-06,
          "val_start": 0.0,
          "warm_up_steps": 100000
        },
        "momentum": {
          "max_decay_steps": 2500000,
          "val_base": 0.994,
          "val_final": 1.0,
          "val_start": 0.0,
          "warm_up_steps": 0
        },
        "teacher_temperature": {
          "max_decay_steps": 37500,
          "val_base": 0.07,
          "val_final": 0.07,
          "val_start": 0.04,
          "warm_up_steps": 37500
        },
        "weight_decay": {
          "max_decay_steps": 2500000,
          "val_base": 0.04,
          "val_final": 0.2,
          "val_start": 0.0,
          "warm_up_steps": 0
        }
      },
      "use_custom_attention": true,
      "validation_interval": 1
    }
  },
  "properties": {
    "automl_disabled_parameters": [
      "wandb",
      "model",
      "dataset",
      "train",
      "inference",
      "export",
      "gen_trt_engine",
      "convert"
    ],
    "convert": {
      "automl_enabled": false,
      "default": {
        "checkpoint": "",
        "output_path": "",
        "results_dir": "",
        "source": "teacher",
        "validate": true
      },
      "description": "Configurable parameters to convert an SSL backbone to the backbone_v2 (timm) layout.",
      "popular": [
        "output_path",
        "source",
        "checkpoint"
      ],
      "properties": {
        "checkpoint": {
          "default": "",
          "description": "SSL DINOv3 checkpoint to convert: a stripped backbone file (student_*.pth / teacher_*.pth) or a full Lightning .pth/.ckpt.",
          "popular": true,
          "title": "checkpoint",
          "type": "string"
        },
        "output_path": {
          "default": "",
          "description": "Output path for the timm-format backbone (.safetensors or .pth). Defaults to <results_dir>/dinov3_<arch>_backbone.safetensors.",
          "popular": true,
          "title": "output path",
          "type": "string"
        },
        "results_dir": {
          "default": "",
          "description": "Directory for convert results/logs",
          "title": "results dir",
          "type": "string"
        },
        "source": {
          "default": "teacher",
          "description": "Which SSL sub-model's backbone to export (EMA teacher recommended).",
          "enum": [
            "student",
            "teacher",
            "student_ema"
          ],
          "popular": true,
          "title": "source",
          "type": "categorical"
        },
        "validate": {
          "default": true,
          "description": "Validate the converted state dict against a fresh timm DINOv3 model.",
          "title": "validate",
          "type": "bool"
        }
      },
      "type": "collection"
    },
    "dataset": {
      "automl_default_parameters": [
        "dataset.batch_size",
        "dataset.workers"
      ],
      "automl_disabled_parameters": [
        "dataset.train_dataset",
        "dataset.test_dataset",
        "dataset.transform"
      ],
      "automl_enabled": false,
      "default": {
        "batch_size": 4,
        "pin_memory": true,
        "test_dataset": {
          "images_dir": ""
        },
        "train_dataset": {
          "images_dir": ""
        },
        "transform": {
          "global_crops_scale": [
            0.32,
            1.0
          ],
          "global_crops_size": 256,
          "local_crops_scale": [
            0.05,
            0.32
          ],
          "local_crops_size": 112,
          "n_global_crops": 2,
          "n_local_crops": 8
        },
        "workers": 8
      },
      "description": "Configurable parameters to construct the dataset for a DINOv3 experiment.",
      "popular": [
        "transform",
        "workers",
        "batch_size",
        "pin_memory"
      ],
      "properties": {
        "batch_size": {
          "automl_enabled": true,
          "default": 4,
          "description": "The batch size for training",
          "maximum": Infinity,
          "minimum": 1,
          "popular": true,
          "title": "batch size",
          "type": "int"
        },
        "pin_memory": {
          "default": true,
          "description": "Flag to enable the dataloader to allocated pagelocked memory for faster of data between the CPU and GPU.",
          "popular": true,
          "title": "pin_memory",
          "type": "bool"
        },
        "test_dataset": {
          "automl_enabled": false,
          "default": {
            "images_dir": ""
          },
          "description": "Configuration for the testing dataset path",
          "properties": {
            "images_dir": {
              "default": "",
              "description": "Path to the extracted images directory for the dataset (not an archive such as .tar.gz)",
              "title": "image directory",
              "type": "string"
            }
          },
          "title": "Testing Dataset",
          "type": "collection"
        },
        "train_dataset": {
          "automl_enabled": false,
          "default": {
            "images_dir": ""
          },
          "description": "Configuration for the training dataset path",
          "properties": {
            "images_dir": {
              "default": "",
              "description": "Path to the extracted images directory for the dataset (not an archive such as .tar.gz)",
              "title": "image directory",
              "type": "string"
            }
          },
          "title": "Training Dataset",
          "type": "collection"
        },
        "transform": {
          "automl_disabled_parameters": [
            "dataset.transform.global_crops_scale",
            "dataset.transform.local_crops_scale"
          ],
          "automl_enabled": false,
          "default": {
            "global_crops_scale": [
              0.32,
              1.0
            ],
            "global_crops_size": 256,
            "local_crops_scale": [
              0.05,
              0.32
            ],
            "local_crops_size": 112,
            "n_global_crops": 2,
            "n_local_crops": 8
          },
          "description": "Configuration parameters for data transformation",
          "popular": [
            "n_global_crops",
            "global_crops_size",
            "local_crops_scale",
            "global_crops_scale",
            "local_crops_size",
            "n_local_crops"
          ],
          "properties": {
            "global_crops_scale": {
              "automl_enabled": false,
              "default": [
                0.32,
                1.0
              ],
              "description": "Scale range for global crops",
              "popular": true,
              "title": "Global Crops Scale",
              "type": "list"
            },
            "global_crops_size": {
              "default": 256,
              "description": "Size of global crops for DINOv3 training.",
              "maximum": Infinity,
              "minimum": 1,
              "popular": true,
              "title": "Global Crops Size",
              "type": "int"
            },
            "local_crops_scale": {
              "automl_enabled": false,
              "default": [
                0.05,
                0.32
              ],
              "description": "Scale range for local crops",
              "popular": true,
              "title": "Local Crops Scale",
              "type": "list"
            },
            "local_crops_size": {
              "default": 112,
              "description": "Size of local crops (multiple of patch 16)",
              "maximum": Infinity,
              "minimum": 1,
              "popular": true,
              "title": "Local Crops Size",
              "type": "int"
            },
            "n_global_crops": {
              "default": 2,
              "description": "Number of global crops to generate",
              "maximum": Infinity,
              "minimum": 1,
              "popular": true,
              "title": "Number of Global Crops",
              "type": "int"
            },
            "n_local_crops": {
              "default": 8,
              "description": "Number of local crops to generate",
              "maximum": Infinity,
              "minimum": 1,
              "popular": true,
              "title": "Number of Local Crops",
              "type": "int"
            }
          },
          "title": "transform",
          "type": "collection"
        },
        "workers": {
          "automl_enabled": true,
          "default": 8,
          "description": "The number of parallel workers processing data",
          "maximum": Infinity,
          "minimum": 1,
          "popular": true,
          "title": "batch size",
          "type": "int"
        }
      },
      "type": "collection"
    },
    "encryption_key": {
      "default": "",
      "description": "Key for encrypting model checkpoints",
      "title": "Encryption key",
      "type": "string"
    },
    "model": {
      "automl_disabled_parameters": [
        "model.distill",
        "model.backbone",
        "model.head",
        "model.gram",
        "model.lora",
        "model.preservation"
      ],
      "automl_enabled": false,
      "default": {
        "backbone": {
          "drop_path_rate": 0.4,
          "img_size": 256,
          "num_register_tokens": 4,
          "patch_size": 16,
          "rope_theta": 100.0,
          "student_type": "vit_b",
          "teacher_type": "vit_b"
        },
        "centering_method": "sinkhorn",
        "distill": {
          "disable_masking": false,
          "enable": false,
          "pretrained_non_distill_pl_model_path": ""
        },
        "gram": {
          "enable": false,
          "refresh_interval": 0,
          "start_step": 0,
          "teacher_scale": 1.0,
          "teacher_source": "pretrained",
          "w_gram": 0.0
        },
        "head": {
          "bottleneck_dim": 384,
          "hidden_dim": 2048,
          "num_layers": 3
        },
        "lora": {
          "alpha": 16.0,
          "dropout": 0.05,
          "enable": false,
          "num_last_blocks": 0,
          "rank": 8,
          "target_modules": [
            "qkv",
            "proj"
          ]
        },
        "preservation": {
          "cls_cosine_weight": 0.05,
          "cls_mse_weight": 0.05,
          "enable": false
        }
      },
      "description": "Configurable parameters to construct the model for a DINOv3 experiment.",
      "popular": [
        "backbone",
        "lora",
        "distill",
        "head",
        "preservation",
        "centering_method",
        "gram"
      ],
      "properties": {
        "backbone": {
          "automl_enabled": false,
          "default": {
            "drop_path_rate": 0.4,
            "img_size": 256,
            "num_register_tokens": 4,
            "patch_size": 16,
            "rope_theta": 100.0,
            "student_type": "vit_b",
            "teacher_type": "vit_b"
          },
          "description": "Configuration for the DINOv3 backbone",
          "popular": [
            "patch_size",
            "rope_theta",
            "img_size",
            "num_register_tokens",
            "drop_path_rate"
          ],
          "properties": {
            "drop_path_rate": {
              "default": 0.4,
              "description": "Drop path rate for stochastic depth regularization",
              "popular": true,
              "title": "drop path rate",
              "type": "float"
            },
            "img_size": {
              "default": 256,
              "description": "Backbone image size. Supported values are 256, 512, and 768.",
              "enum": [
                256,
                512,
                768
              ],
              "popular": true,
              "title": "image size",
              "type": "ordered_int"
            },
            "num_register_tokens": {
              "default": 4,
              "description": "Number of register tokens (DINOv3 ViT-B uses 4)",
              "maximum": Infinity,
              "minimum": 0,
              "popular": true,
              "title": "num register tokens",
              "type": "int"
            },
            "patch_size": {
              "default": 16,
              "description": "Size of patches (DINOv3 uses patch 16)",
              "enum": [
                16
              ],
              "popular": true,
              "title": "patch size",
              "type": "ordered_int"
            },
            "rope_theta": {
              "default": 100.0,
              "description": "Frequency base for 2D axial RoPE. Must match the timm DINOv3 reference; verified by the step-4 feature-parity smoke test.",
              "popular": true,
              "title": "RoPE theta",
              "type": "float"
            },
            "student_type": {
              "default": "vit_b",
              "description": "Student backbone name. TAO's DINOv3 supports vit_s, vit_s_plus, vit_b, vit_l, vit_h_plus and vit_7b.",
              "enum": [
                "vit_s",
                "vit_s_plus",
                "vit_b",
                "vit_l",
                "vit_h_plus",
                "vit_7b"
              ],
              "title": "student backbone",
              "type": "categorical"
            },
            "teacher_type": {
              "default": "vit_b",
              "description": "Teacher backbone name. TAO's DINOv3 supports vit_s, vit_s_plus, vit_b, vit_l, vit_h_plus and vit_7b.",
              "enum": [
                "vit_s",
                "vit_s_plus",
                "vit_b",
                "vit_l",
                "vit_h_plus",
                "vit_7b"
              ],
              "title": "teacher backbone",
              "type": "categorical"
            }
          },
          "type": "collection"
        },
        "centering_method": {
          "default": "sinkhorn",
          "description": "Teacher-output centering for the DINO/iBOT heads. DINOv3 uses Sinkhorn-Knopp (SwAV); 'softmax' is the DINOv2 fallback. If training shows instability/collapse, try 'softmax'.",
          "enum": [
            "sinkhorn",
            "softmax"
          ],
          "popular": true,
          "title": "centering method",
          "type": "categorical"
        },
        "distill": {
          "automl_enabled": false,
          "default": {
            "disable_masking": false,
            "enable": false,
            "pretrained_non_distill_pl_model_path": ""
          },
          "description": "Configuration for distillation (reused from nvdinov2)",
          "popular": [
            "enable",
            "disable_masking"
          ],
          "properties": {
            "disable_masking": {
              "default": false,
              "description": "Whether to disable masking when distillation",
              "popular": true,
              "title": "disable_masking",
              "type": "bool"
            },
            "enable": {
              "default": false,
              "description": "Whether to run distillation",
              "popular": true,
              "title": "distillation",
              "type": "bool"
            },
            "pretrained_non_distill_pl_model_path": {
              "default": "",
              "description": "Path to a pre-trained pl model from non-distillation DINOv2 SSL pipe for initializing teacher in distillation.",
              "type": "string"
            }
          },
          "type": "collection"
        },
        "gram": {
          "automl_enabled": false,
          "default": {
            "enable": false,
            "refresh_interval": 0,
            "start_step": 0,
            "teacher_scale": 1.0,
            "teacher_source": "pretrained",
            "w_gram": 0.0
          },
          "description": "Configuration for DINOv3 Gram anchoring",
          "popular": [
            "start_step",
            "enable",
            "w_gram"
          ],
          "properties": {
            "enable": {
              "default": false,
              "description": "Whether to add the Gram-anchoring loss term",
              "popular": true,
              "title": "enable gram",
              "type": "bool"
            },
            "refresh_interval": {
              "default": 0,
              "description": "Steps between refreshing the Gram teacher from the EMA teacher (only when teacher_source='ema'). 0 = never refresh (frozen snapshot, Phase 0 behavior).",
              "maximum": Infinity,
              "minimum": 0,
              "title": "gram refresh interval",
              "type": "int"
            },
            "start_step": {
              "default": 0,
              "description": "Global step at which the Gram term activates",
              "maximum": Infinity,
              "minimum": 0,
              "popular": true,
              "title": "gram start step",
              "type": "int"
            },
            "teacher_scale": {
              "default": 1.0,
              "description": "Resolution multiple at which the Gram teacher runs relative to the student. DINOv3 computes Gram on higher-res teacher features; the teacher grid is average-pooled back to the student grid before the loss. 1.0 = same resolution (default, memory-safe). The paper uses 2.0, but at 768 the teacher then runs at 1536 (96x96=9216 tokens) -> heavy; raise to 1.5/2.0 only if memory allows.",
              "title": "gram teacher scale",
              "type": "float"
            },
            "teacher_source": {
              "default": "pretrained",
              "description": "Source of the frozen Gram teacher weights: 'pretrained' anchors to the loaded DINOv3 weights (Phase 0 default); 'ema' uses an early-EMA snapshot of the current run's teacher, refreshed every 'refresh_interval' steps (Phase 1 / high-res).",
              "enum": [
                "pretrained",
                "ema"
              ],
              "title": "gram teacher source",
              "type": "categorical"
            },
            "w_gram": {
              "default": 0.0,
              "description": "Weight of the Gram-anchoring loss term",
              "popular": true,
              "title": "gram weight",
              "type": "float"
            }
          },
          "type": "collection"
        },
        "head": {
          "automl_enabled": false,
          "default": {
            "bottleneck_dim": 384,
            "hidden_dim": 2048,
            "num_layers": 3
          },
          "description": "Configuration for the DINOv3 head (reused from nvdinov2)",
          "popular": [
            "bottleneck_dim",
            "num_layers",
            "hidden_dim"
          ],
          "properties": {
            "bottleneck_dim": {
              "default": 384,
              "description": "Dimension of the bottleneck layer in the NVDINOv2 head",
              "maximum": Infinity,
              "minimum": 1,
              "popular": true,
              "title": "bottleneck dimension",
              "type": "int"
            },
            "hidden_dim": {
              "default": 2048,
              "description": "Dimension of the hidden layers in the NVDINOv2 head",
              "maximum": Infinity,
              "minimum": 1,
              "popular": true,
              "title": "hidden dimension",
              "type": "int"
            },
            "num_layers": {
              "default": 3,
              "description": "Number of layers in the NVDINOv2 head",
              "maximum": Infinity,
              "minimum": 1,
              "popular": true,
              "title": "number of Layers",
              "type": "int"
            }
          },
          "type": "collection"
        },
        "lora": {
          "automl_disabled_parameters": [
            "model.lora.target_modules"
          ],
          "automl_enabled": false,
          "default": {
            "alpha": 16.0,
            "dropout": 0.05,
            "enable": false,
            "num_last_blocks": 0,
            "rank": 8,
            "target_modules": [
              "qkv",
              "proj"
            ]
          },
          "description": "Configuration for LoRA parameter-efficient continual pre-training",
          "popular": [
            "rank",
            "num_last_blocks",
            "alpha",
            "target_modules",
            "enable"
          ],
          "properties": {
            "alpha": {
              "default": 16.0,
              "description": "LoRA scaling alpha; the applied scale is alpha / rank.",
              "popular": true,
              "title": "lora alpha",
              "type": "float"
            },
            "dropout": {
              "default": 0.05,
              "description": "Dropout applied to the LoRA branch input only (the frozen base path is untouched).",
              "maximum": 1.0,
              "minimum": 0.0,
              "title": "lora dropout",
              "type": "float"
            },
            "enable": {
              "default": false,
              "description": "Whether to apply LoRA adapters to the backbone (parameter-efficient SSL)",
              "popular": true,
              "title": "enable lora",
              "type": "bool"
            },
            "num_last_blocks": {
              "default": 0,
              "description": "Adapt only the last N transformer blocks. 0 = all blocks.",
              "maximum": Infinity,
              "minimum": 0,
              "popular": true,
              "title": "lora num last blocks",
              "type": "int"
            },
            "rank": {
              "default": 8,
              "description": "LoRA rank (r). The low-rank update is dW = (alpha/r) * B @ A.",
              "maximum": Infinity,
              "minimum": 1,
              "popular": true,
              "title": "lora rank",
              "type": "int"
            },
            "target_modules": {
              "automl_enabled": false,
              "default": [
                "qkv",
                "proj"
              ],
              "description": "Projections to adapt inside each targeted ViT block. 'qkv' and 'proj' are the attention projections (the v1 default); 'fc1'/'fc2' additionally adapt the FFN. SwiGLU backbones (ViT-S+/H+/7B) expose a fused fc1 and are deferred in v1.",
              "popular": true,
              "title": "lora target modules",
              "type": "list"
            }
          },
          "type": "collection"
        },
        "preservation": {
          "automl_enabled": false,
          "default": {
            "cls_cosine_weight": 0.05,
            "cls_mse_weight": 0.05,
            "enable": false
          },
          "description": "Configuration for CLS-token preservation against the frozen anchor teacher",
          "popular": [
            "cls_mse_weight",
            "cls_cosine_weight",
            "enable"
          ],
          "properties": {
            "cls_cosine_weight": {
              "default": 0.05,
              "description": "Weight of the CLS-token cosine preservation term (0 disables just this term).",
              "maximum": Infinity,
              "minimum": 0.0,
              "popular": true,
              "title": "cls cosine weight",
              "type": "float"
            },
            "cls_mse_weight": {
              "default": 0.05,
              "description": "Weight of the CLS-token MSE preservation term (0 disables just this term).",
              "maximum": Infinity,
              "minimum": 0.0,
              "popular": true,
              "title": "cls mse weight",
              "type": "float"
            },
            "enable": {
              "default": false,
              "description": "Whether to add CLS-token preservation losses against the frozen anchor teacher. Recommended alongside LoRA for continual pre-training.",
              "popular": true,
              "title": "enable preservation",
              "type": "bool"
            }
          },
          "type": "collection"
        }
      },
      "type": "collection"
    },
    "model_name": {
      "default": "",
      "description": "Name of model if invoking task via :code:`model_agnostic`",
      "title": "Model name",
      "type": "string"
    },
    "results_dir": {
      "default": "",
      "description": "\n        Path to where all the assets generated from a task are stored.\n        ",
      "title": "Results directory",
      "type": "string"
    },
    "train": {
      "automl_disabled_parameters": [
        "train.gpu_ids",
        "train.cudnn",
        "train.checkpointer",
        "train.schedulers",
        "train.optim"
      ],
      "automl_enabled": false,
      "default": {
        "checkpoint_interval": 1,
        "checkpoint_interval_unit": "epoch",
        "checkpointer": {
          "auto_insert_metric_name": false,
          "enable_topk": false,
          "filename": "model_best_{epoch:03d}",
          "replace_periodic": false,
          "save_top_k": 1
        },
        "clip_grad_norm": 3.0,
        "cudnn": {
          "benchmark": true,
          "deterministic": false
        },
        "distributed_strategy": "auto",
        "gpu_ids": [
          0
        ],
        "layerwise_decay": 1.0,
        "log_every_n_steps": 1,
        "num_epochs": 10,
        "num_gpus": 1,
        "num_nodes": 1,
        "num_prototypes": 131072,
        "optim": {
          "optim": "adamw"
        },
        "precision": "16-mixed",
        "results_dir": "",
        "resume_training_checkpoint_path": "",
        "schedulers": {
          "last_layer_learning_rate": {
            "freeze_steps": 1250,
            "max_decay_steps": 2500000,
            "val_base": 7.07e-06,
            "val_final": 1e-06,
            "val_start": 0.0,
            "warm_up_steps": 100000
          },
          "learning_rate": {
            "max_decay_steps": 2500000,
            "val_base": 7.07e-06,
            "val_final": 1e-06,
            "val_start": 0.0,
            "warm_up_steps": 100000
          },
          "momentum": {
            "max_decay_steps": 2500000,
            "val_base": 0.994,
            "val_final": 1.0,
            "val_start": 0.0,
            "warm_up_steps": 0
          },
          "teacher_temperature": {
            "max_decay_steps": 37500,
            "val_base": 0.07,
            "val_final": 0.07,
            "val_start": 0.04,
            "warm_up_steps": 37500
          },
          "weight_decay": {
            "max_decay_steps": 2500000,
            "val_base": 0.04,
            "val_final": 0.2,
            "val_start": 0.0,
            "warm_up_steps": 0
          }
        },
        "seed": 1234,
        "use_custom_attention": true,
        "validation_interval": 1
      },
      "description": "Configurable parameters to construct the trainer for a DINOv3 experiment.",
      "popular": [
        "layerwise_decay",
        "checkpoint_interval",
        "optim",
        "gpu_ids",
        "validation_interval",
        "num_gpus",
        "precision",
        "num_prototypes",
        "schedulers",
        "distributed_strategy",
        "num_nodes",
        "clip_grad_norm",
        "num_epochs",
        "use_custom_attention"
      ],
      "properties": {
        "checkpoint_interval": {
          "default": 1,
          "description": "The interval (in epochs) at which a checkpoint will be saved. Helps resume training.",
          "minimum": 1,
          "popular": true,
          "title": "Checkpoint interval",
          "type": "int"
        },
        "checkpoint_interval_unit": {
          "default": "epoch",
          "description": "The unit of the checkpoint interval.",
          "enum": [
            "epoch",
            "step"
          ],
          "title": "Checkpoint interval unit",
          "type": "categorical"
        },
        "checkpointer": {
          "automl_enabled": false,
          "default": {
            "auto_insert_metric_name": false,
            "enable_topk": false,
            "filename": "model_best_{epoch:03d}",
            "replace_periodic": false,
            "save_top_k": 1
          },
          "properties": {
            "auto_insert_metric_name": {
              "default": false,
              "type": "bool"
            },
            "dirpath": {
              "description": "Directory for best checkpoints. Defaults to results_dir.",
              "type": "string"
            },
            "enable_topk": {
              "default": false,
              "description": "Save checkpoint(s) ranked by a monitored metric. This is additive to periodic checkpoints unless replace_periodic is enabled.",
              "title": "Enable best-checkpoint saving",
              "type": "bool"
            },
            "filename": {
              "default": "model_best_{epoch:03d}",
              "title": "Best-checkpoint filename pattern",
              "type": "string"
            },
            "mode": {
              "description": "Leave unset to use the network's default direction.",
              "enum": [
                "min",
                "max"
              ],
              "title": "Monitor mode",
              "type": "categorical"
            },
            "monitor": {
              "description": "Leave unset to use the network's default metric.",
              "title": "Monitored metric",
              "type": "string"
            },
            "replace_periodic": {
              "default": false,
              "description": "When best-checkpoint saving is enabled, replace periodic history with one metric-best checkpoint and its latest symlink.",
              "title": "Replace periodic checkpoints",
              "type": "bool"
            },
            "save_top_k": {
              "default": 1,
              "minimum": 1,
              "title": "Number of best checkpoints",
              "type": "int"
            }
          },
          "type": "collection"
        },
        "clip_grad_norm": {
          "default": 3.0,
          "description": "Value to clip gradients norm",
          "popular": true,
          "title": "clip gradient norm",
          "type": "float"
        },
        "cudnn": {
          "automl_enabled": false,
          "default": {
            "benchmark": true,
            "deterministic": false
          },
          "description": "CuDNN settings compatible with DINOv3 custom attention.",
          "properties": {
            "benchmark": {
              "default": true,
              "description": "Enable CuDNN benchmarking for DINOv3 training.",
              "title": "CuDNN benchmark",
              "type": "bool"
            },
            "deterministic": {
              "default": false,
              "description": "Enable deterministic CuDNN behavior. Keep disabled when using DINOv3 custom attention, which has no deterministic backward implementation.",
              "title": "CuDNN deterministic",
              "type": "bool"
            }
          },
          "type": "collection"
        },
        "distributed_strategy": {
          "default": "auto",
          "description": "Lightning distributed strategy. 'auto' keeps the nvdinov2 behaviour (single-device or DDP). 'fsdp' shards params/grads/optimizer (FULL_SHARD) for high-resolution and large-backbone multi-GPU training.",
          "enum": [
            "auto",
            "ddp",
            "fsdp"
          ],
          "popular": true,
          "title": "distributed strategy",
          "type": "categorical"
        },
        "gpu_ids": {
          "automl_enabled": false,
          "default": [
            0
          ],
          "description": "\n        List of GPU IDs to run the training on. The length of this list\n        must be equal to the number of gpus in train.num_gpus.",
          "popular": true,
          "title": "GPU IDs",
          "type": "list"
        },
        "layerwise_decay": {
          "default": 1.0,
          "description": "Layerwise decay factor",
          "popular": true,
          "title": "layerwise decay factor",
          "type": "float"
        },
        "log_every_n_steps": {
          "default": 1,
          "description": "Number of training steps between logger updates.",
          "minimum": 1,
          "title": "logging interval",
          "type": "int"
        },
        "num_epochs": {
          "default": 10,
          "description": "Number of epochs to run the training.",
          "maximum": Infinity,
          "minimum": 1,
          "popular": true,
          "title": "Number of epochs",
          "type": "int"
        },
        "num_gpus": {
          "default": 1,
          "description": "The number of GPUs to run the train job.",
          "minimum": 1,
          "popular": true,
          "title": "Number of GPUs",
          "type": "int"
        },
        "num_nodes": {
          "default": 1,
          "description": "Number of nodes to run the training on. If > 1, then multi-node is enabled.",
          "minimum": 1,
          "popular": true,
          "title": "Number of nodes",
          "type": "int"
        },
        "num_prototypes": {
          "default": 131072,
          "description": "Number of prototypes",
          "popular": true,
          "title": "number of prototypes",
          "type": "int"
        },
        "optim": {
          "automl_enabled": false,
          "default": {
            "optim": "adamw"
          },
          "description": "Optimizer configuration for NVDINOv2",
          "popular": [
            "optim"
          ],
          "properties": {
            "optim": {
              "default": "adamw",
              "description": "Optimizer type",
              "enum": [
                "adamw",
                ""
              ],
              "popular": true,
              "title": "optimizer",
              "type": "categorical"
            }
          },
          "type": "collection"
        },
        "precision": {
          "default": "16-mixed",
          "description": "Precision",
          "popular": true,
          "title": "precision",
          "type": "string"
        },
        "pretrained_model_path": {
          "description": "Path to DINOv3 pretrained weights matching the configured backbone. Accepts a timm-format directory or file, or a stripped TAO DINOv3 backbone checkpoint. DINOv2/NVDINOv2 checkpoints are not supported.",
          "type": "string"
        },
        "results_dir": {
          "default": "",
          "description": "\n        Path to where all the assets generated from a task are stored.\n        ",
          "title": "Results directory",
          "type": "string"
        },
        "resume_training_checkpoint_path": {
          "default": "",
          "description": "Path to the checkpoint to resume training from.",
          "title": "Resume checkpoint path",
          "type": "string"
        },
        "schedulers": {
          "automl_disabled_parameters": [
            "train.schedulers.learning_rate",
            "train.schedulers.last_layer_learning_rate",
            "train.schedulers.weight_decay",
            "train.schedulers.momentum",
            "train.schedulers.teacher_temperature"
          ],
          "automl_enabled": false,
          "default": {
            "last_layer_learning_rate": {
              "freeze_steps": 1250,
              "max_decay_steps": 2500000,
              "val_base": 7.07e-06,
              "val_final": 1e-06,
              "val_start": 0.0,
              "warm_up_steps": 100000
            },
            "learning_rate": {
              "max_decay_steps": 2500000,
              "val_base": 7.07e-06,
              "val_final": 1e-06,
              "val_start": 0.0,
              "warm_up_steps": 100000
            },
            "momentum": {
              "max_decay_steps": 2500000,
              "val_base": 0.994,
              "val_final": 1.0,
              "val_start": 0.0,
              "warm_up_steps": 0
            },
            "teacher_temperature": {
              "max_decay_steps": 37500,
              "val_base": 0.07,
              "val_final": 0.07,
              "val_start": 0.04,
              "warm_up_steps": 37500
            },
            "weight_decay": {
              "max_decay_steps": 2500000,
              "val_base": 0.04,
              "val_final": 0.2,
              "val_start": 0.0,
              "warm_up_steps": 0
            }
          },
          "description": "Schedulers configuration for NVDINOv2 training",
          "popular": [
            "learning_rate",
            "weight_decay",
            "teacher_temperature",
            "momentum",
            "last_layer_learning_rate"
          ],
          "properties": {
            "last_layer_learning_rate": {
              "automl_enabled": false,
              "default": {
                "freeze_steps": 1250,
                "max_decay_steps": 2500000,
                "val_base": 7.07e-06,
                "val_final": 1e-06,
                "val_start": 0.0,
                "warm_up_steps": 100000
              },
              "description": "Last layer learning rate scheduler configuration",
              "popular": [
                "val_base",
                "val_final",
                "val_start",
                "warm_up_steps",
                "freeze_steps",
                "max_decay_steps"
              ],
              "properties": {
                "freeze_steps": {
                  "default": 1250,
                  "description": "Number of freeze steps",
                  "popular": true,
                  "title": "freeze steps",
                  "type": "int"
                },
                "max_decay_steps": {
                  "default": 2500000,
                  "description": "Maximum decay steps",
                  "popular": true,
                  "title": "max decay steps",
                  "type": "int"
                },
                "val_base": {
                  "default": 7.07e-06,
                  "description": "The value after warm-up for scheduler.",
                  "popular": true,
                  "title": "base value",
                  "type": "float"
                },
                "val_final": {
                  "default": 1e-06,
                  "description": "Final value for scheduler",
                  "popular": true,
                  "title": "final value",
                  "type": "float"
                },
                "val_start": {
                  "default": 0.0,
                  "description": "Starting value for scheduler",
                  "popular": true,
                  "title": "starting value",
                  "type": "float"
                },
                "warm_up_steps": {
                  "default": 100000,
                  "description": "Number of warm-up steps",
                  "popular": true,
                  "title": "warm-up steps",
                  "type": "int"
                }
              },
              "type": "collection"
            },
            "learning_rate": {
              "automl_enabled": false,
              "default": {
                "max_decay_steps": 2500000,
                "val_base": 7.07e-06,
                "val_final": 1e-06,
                "val_start": 0.0,
                "warm_up_steps": 100000
              },
              "description": "Learning rate scheduler configuration",
              "popular": [
                "val_base",
                "val_final",
                "val_start",
                "warm_up_steps",
                "max_decay_steps"
              ],
              "properties": {
                "max_decay_steps": {
                  "default": 2500000,
                  "description": "Maximum decay steps",
                  "popular": true,
                  "title": "max decay steps",
                  "type": "int"
                },
                "val_base": {
                  "default": 7.07e-06,
                  "description": "The value after warm-up for scheduler",
                  "popular": true,
                  "title": "base value",
                  "type": "float"
                },
                "val_final": {
                  "default": 1e-06,
                  "description": "Final value for scheduler",
                  "popular": true,
                  "title": "final value",
                  "type": "float"
                },
                "val_start": {
                  "default": 0.0,
                  "description": "Starting value for scheduler",
                  "popular": true,
                  "title": "starting value",
                  "type": "float"
                },
                "warm_up_steps": {
                  "default": 100000,
                  "description": "Number of warm-up steps",
                  "popular": true,
                  "title": "warm-up steps",
                  "type": "int"
                }
              },
              "type": "collection"
            },
            "momentum": {
              "automl_enabled": false,
              "default": {
                "max_decay_steps": 2500000,
                "val_base": 0.994,
                "val_final": 1.0,
                "val_start": 0.0,
                "warm_up_steps": 0
              },
              "description": "Momentum scheduler configuration",
              "popular": [
                "val_base",
                "val_final",
                "val_start",
                "warm_up_steps",
                "max_decay_steps"
              ],
              "properties": {
                "max_decay_steps": {
                  "default": 2500000,
                  "description": "Maximum decay steps",
                  "popular": true,
                  "title": "max decay steps",
                  "type": "int"
                },
                "val_base": {
                  "default": 0.994,
                  "description": "The value after warm-up for scheduler",
                  "popular": true,
                  "title": "base value",
                  "type": "float"
                },
                "val_final": {
                  "default": 1.0,
                  "description": "Final value for scheduler",
                  "popular": true,
                  "title": "final value",
                  "type": "float"
                },
                "val_start": {
                  "default": 0.0,
                  "description": "Starting value for scheduler",
                  "popular": true,
                  "title": "starting value",
                  "type": "float"
                },
                "warm_up_steps": {
                  "default": 0,
                  "description": "Number of warm-up steps",
                  "popular": true,
                  "title": "warm-up steps",
                  "type": "int"
                }
              },
              "type": "collection"
            },
            "teacher_temperature": {
              "automl_enabled": false,
              "default": {
                "max_decay_steps": 37500,
                "val_base": 0.07,
                "val_final": 0.07,
                "val_start": 0.04,
                "warm_up_steps": 37500
              },
              "description": "Teacher temperature scheduler configuration",
              "popular": [
                "val_base",
                "val_final",
                "val_start",
                "warm_up_steps",
                "max_decay_steps"
              ],
              "properties": {
                "max_decay_steps": {
                  "default": 37500,
                  "description": "Maximum decay steps",
                  "popular": true,
                  "title": "max decay steps",
                  "type": "int"
                },
                "val_base": {
                  "default": 0.07,
                  "description": "The value after warm-up for scheduler",
                  "popular": true,
                  "title": "base value",
                  "type": "float"
                },
                "val_final": {
                  "default": 0.07,
                  "description": "Final value for scheduler",
                  "popular": true,
                  "title": "final value",
                  "type": "float"
                },
                "val_start": {
                  "default": 0.04,
                  "description": "Starting value for scheduler",
                  "popular": true,
                  "title": "starting value",
                  "type": "float"
                },
                "warm_up_steps": {
                  "default": 37500,
                  "description": "Number of warm-up steps",
                  "popular": true,
                  "title": "warm-up steps",
                  "type": "int"
                }
              },
              "type": "collection"
            },
            "weight_decay": {
              "automl_enabled": false,
              "default": {
                "max_decay_steps": 2500000,
                "val_base": 0.04,
                "val_final": 0.2,
                "val_start": 0.0,
                "warm_up_steps": 0
              },
              "description": "Weight decay scheduler configuration",
              "popular": [
                "val_base",
                "val_final",
                "val_start",
                "warm_up_steps",
                "max_decay_steps"
              ],
              "properties": {
                "max_decay_steps": {
                  "default": 2500000,
                  "description": "Maximum decay steps",
                  "popular": true,
                  "title": "max decay steps",
                  "type": "int"
                },
                "val_base": {
                  "default": 0.04,
                  "description": "The value after warm-up for scheduler",
                  "popular": true,
                  "title": "base value",
                  "type": "float"
                },
                "val_final": {
                  "default": 0.2,
                  "description": "Final value for scheduler",
                  "popular": true,
                  "title": "final value",
                  "type": "float"
                },
                "val_start": {
                  "default": 0.0,
                  "description": "Starting value for scheduler",
                  "popular": true,
                  "title": "starting value",
                  "type": "float"
                },
                "warm_up_steps": {
                  "default": 0,
                  "description": "Number of warm-up steps",
                  "popular": true,
                  "title": "warm-up steps",
                  "type": "int"
                }
              },
              "type": "collection"
            }
          },
          "type": "collection"
        },
        "seed": {
          "default": 1234,
          "description": "The seed for the initializer in PyTorch. If < 0, disable fixed seed.",
          "maximum": Infinity,
          "minimum": -1,
          "title": "Seed for randomization",
          "type": "int"
        },
        "use_custom_attention": {
          "default": true,
          "description": "Whether to use memory_efficient_attention",
          "popular": true,
          "title": "custom_attention",
          "type": "bool"
        },
        "validation_interval": {
          "default": 1,
          "description": "\n        The interval (in epochs) at which a evaluation\n        will be triggered on the validation dataset.",
          "minimum": 1,
          "popular": true,
          "title": "Validation interval",
          "type": "int"
        }
      },
      "type": "collection"
    },
    "wandb": {
      "automl_disabled_parameters": [
        "wandb.tags"
      ],
      "automl_enabled": false,
      "default": {
        "enable": true,
        "entity": "",
        "group": "",
        "name": "TAO Toolkit Training",
        "project": "TAO Toolkit",
        "reinit": false,
        "run_id": "",
        "save_code": false,
        "sync_tensorboard": false,
        "tags": [
          "tao-toolkit"
        ]
      },
      "properties": {
        "enable": {
          "default": true,
          "type": "bool"
        },
        "entity": {
          "default": "",
          "type": "string"
        },
        "group": {
          "default": "",
          "type": "string"
        },
        "name": {
          "default": "TAO Toolkit Training",
          "type": "string"
        },
        "project": {
          "default": "TAO Toolkit",
          "type": "string"
        },
        "reinit": {
          "default": false,
          "type": "bool"
        },
        "run_id": {
          "default": "",
          "type": "string"
        },
        "save_code": {
          "default": false,
          "type": "bool"
        },
        "sync_tensorboard": {
          "default": false,
          "type": "bool"
        },
        "tags": {
          "automl_enabled": false,
          "default": [
            "tao-toolkit"
          ],
          "type": "list"
        }
      },
      "type": "collection"
    }
  },
  "type": "object",
  "x_tao_schema": {
    "action": "convert",
    "core_module": "dinov3",
    "model": "tao-train-dinov3",
    "network_arch": "dinov3",
    "schema_action": "convert",
    "schema_version": 1,
    "source": "tao-core dataclass config"
  }
}

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Needs: Declares allowed tools, Needs a command-line tool

The author says: Requires docker + nvidia-container-toolkit.

Security

What our scanner noticed· Medium risk

Large base64-encoded blobs

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Created by

NVIDIA Corporation

Publisher

NVIDIA (Official)

Source

NVIDIA/skills · skills/tao-train-dinov3 · commit fce3cca

Licence

Apache License 2.0 (redistribution allowed)

LICENSE (added by AutoGineer from LICENSE-APACHE)
ATTRIBUTION.md (written by AutoGineer)

Synced

2026-11-08

AutoGineer is not affiliated with or endorsed by NVIDIA Corporation or NVIDIA.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download tao-train-dinov3.zip