tao-train-nvdinov2

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

NVDINOv2

NVDINOv2 for self-supervised visual representation learning. Trains vision transformers via self-distillation (teacher-student) without labels. Produces general-purpose visual features.

Set train.pretrained_model_path for pretrained ViT weights.

For TAO Deploy TensorRT actions (

gen_trt_engine

), read

references/tao-deploy-nvdinov2.md

first. Deploy spec templates live in this skill's

references/

folder with the

spec_template_deploy_*.yaml

prefix.

NVDINOv2用于自监督视觉表示学习。通过无标签的自蒸馏（师生模型）方式训练Vision Transformer，生成通用视觉特征。

设置train.pretrained_model_path以加载预训练ViT权重。

对于TAO Deploy TensorRT操作（

gen_trt_engine

），请先阅读

references/tao-deploy-nvdinov2.md

。部署规格模板存放在本skill的

references/

文件夹中，前缀为

spec_template_deploy_*.yaml

。

Dataclass Schemas

数据类模式

Generated TAO Core schemas are packaged in

schemas/<action>.schema.json

, with

schemas/manifest.json

listing available actions. Each generated schema also emits

references/spec_template_<action>.yaml

from the schema top-level

default

field. AutoML enablement is declared at the model layer in

references/skill_info.yaml

via

automl_enabled

. Runnable AutoML still requires

schemas/train.schema.json

and

references/spec_template_train.yaml

to exist and parse. Use the packaged train schema for

automl_default_parameters

automl_disabled_parameters

, defaults, min/max bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect

~/tao-core

at runtime; maintainers regenerate schemas/templates before packaging the skill bank.

生成的TAO Core模式打包在

schemas/<action>.schema.json

中，

schemas/manifest.json

列出了可用操作。每个生成的模式还会从模式顶层的

default

字段生成

references/spec_template_<action>.yaml

。AutoML支持在

references/skill_info.yaml

的模型层通过

automl_enabled

声明。可运行的AutoML仍要求

schemas/train.schema.json

和

references/spec_template_train.yaml

存在且可解析。使用打包的训练模式来配置

automl_default_parameters

、

automl_disabled_parameters

、默认值、最小/最大范围、枚举值、选项权重、数学条件、依赖关系以及常用参数。运行时不要依赖

~/tao-core

；维护人员在打包skill库前会重新生成模式/模板。

Train Action Policy

训练操作策略

This model is AutoML-enabled at the model layer. Before handling any train-stage request, read

references/skill_info.yaml

and resolve the run override from either an explicit

automl_policy

value or the user's workflow request. Treat phrases like "turn off AutoML", "disable AutoML", "no HPO", or "plain training" as

automl_policy: off

for this run only; otherwise default to

auto

. When

automl_policy: auto

automl_enabled: true

, and both

schemas/train.schema.json

and

references/spec_template_train.yaml

are packaged, route the train action through

tao-skill-bank:tao-run-automl

by default with this model's

skill_dir

. Preserve workflow/application overrides for datasets, specs, output directories, GPU/platform settings, parent checkpoints, and

automl_policy

. Use direct model training only when

automl_policy: off

or the packaged train schema/template is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.

Non-train actions such as

evaluate

inference

export

, and deploy flows stay in this model skill. The per-run

automl_policy

override does not change model metadata.

该模型在模型层支持AutoML。处理任何训练阶段请求前，请先阅读

references/skill_info.yaml

，并通过显式的

automl_policy

值或用户的工作流请求确定运行覆盖配置。将“turn off AutoML”、“disable AutoML”、“no HPO”或“plain training”这类短语视为本次运行的

automl_policy: off

；否则默认设为

auto

。当

automl_policy: auto

、

automl_enabled: true

，且

schemas/train.schema.json

和

references/spec_template_train.yaml

已打包时，默认将训练操作通过

tao-skill-bank:tao-run-automl

路由，并传入该模型的

skill_dir

。保留数据集、规格、输出目录、GPU/平台设置、父检查点和

automl_policy

的工作流/应用覆盖配置。仅当

automl_policy: off

或打包的训练模式/模板缺失时，才使用直接模型训练；若模式缺失，需报告该模型已启用AutoML但无法运行，直到生成模式为止。

非训练操作（如

evaluate

、

inference

、

export

以及部署流程）仍在本模型skill中执行。每次运行的

automl_policy

覆盖配置不会更改模型元数据。

Training Requirements

训练要求

Dataset type: image_classification
Formats: ssl
Monitoring metric: train_loss

数据集类型： image_classification
格式： ssl
监控指标： train_loss

Per-Action Dataset Requirements

各操作的数据集要求

Action	Spec Key	Source	Files	List?
distill	dataset.train_dataset.images_dir	train_datasets	images_train.tar.gz	No
inference	dataset.test_dataset.images_dir	inference_dataset	images_test.tar.gz	No
train	dataset.train_dataset.images_dir	train_datasets	images_train.tar.gz	No

操作	规格键	来源	文件	是否为列表？
distill	dataset.train_dataset.images_dir	train_datasets	images_train.tar.gz	否
inference	dataset.test_dataset.images_dir	inference_dataset	images_test.tar.gz	否
train	dataset.train_dataset.images_dir	train_datasets	images_train.tar.gz	否

Typical Spec Overrides

典型规格覆盖配置

Data source overrides are mandatory for every action — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in

spec_overrides

python

S3_TRAIN = "s3://bucket/data/train"
S3_EVAL = "s3://bucket/data/eval"

train (mandatory data sources):

python

{
    "train.num_gpus": 1,
    "train.num_epochs": 10,
    "train.checkpoint_interval": 10,
    "dataset.train_dataset.images_dir": f"{S3_TRAIN}/images_train.tar.gz",
}

distill (mandatory data sources):

python

{
    "dataset.train_dataset.images_dir": f"{S3_TRAIN}/images_train.tar.gz",
}

inference (mandatory data sources):

python

{
    "dataset.test_dataset.images_dir": f"{S3_EVAL}/images_test.tar.gz",
}

数据源覆盖配置对每个操作都是必填项——Agent必须根据上述“各操作的数据集要求”表格构建数据源路径，并将其包含在

spec_overrides

中。

python

S3_TRAIN = "s3://bucket/data/train"
S3_EVAL = "s3://bucket/data/eval"

训练（必填数据源）：

python

{
    "train.num_gpus": 1,
    "train.num_epochs": 10,
    "train.checkpoint_interval": 10,
    "dataset.train_dataset.images_dir": f"{S3_TRAIN}/images_train.tar.gz",
}

蒸馏（必填数据源）：

python

{
    "dataset.train_dataset.images_dir": f"{S3_TRAIN}/images_train.tar.gz",
}

推理（必填数据源）：

python

{
    "dataset.test_dataset.images_dir": f"{S3_EVAL}/images_test.tar.gz",
}

Eval Dataset

评估数据集

Optional. SSL training does not use labels. Evaluation is downstream task-specific.

可选。SSL训练不使用标签。评估针对下游特定任务。

Important Parameters

重要参数

model.backbone.teacher_type: Teacher ViT variant. Default vit_l (ViT-Large).
model.backbone.student_type: Student ViT variant. Default vit_l. Typically matches teacher.
model.backbone.img_size: Input image size. Default 518. Higher resolution produces better features but costs more memory.
model.backbone.patch_size: ViT patch size. Default 14.
dataset.batch_size: Per-GPU batch size. Default 4. SSL training is memory-intensive due to dual (teacher+student) forward passes.
train.layerwise_decay: Layer-wise learning rate decay. Important for ViT fine-tuning.
train.clip_grad_norm: Gradient clipping. Important for stable SSL training.

model.backbone.teacher_type：教师ViT变体。默认值为vit_l（ViT-Large）。
model.backbone.student_type：学生ViT变体。默认值为vit_l。通常与教师模型匹配。
model.backbone.img_size：输入图像尺寸。默认值为518。更高分辨率会生成更好的特征，但会占用更多内存。
model.backbone.patch_size：ViT补丁尺寸。默认值为14。
dataset.batch_size：单GPU批次大小。默认值为4。由于需要执行（教师+学生）双前向传播，SSL训练对内存要求很高。
train.layerwise_decay：分层学习率衰减。对ViT微调至关重要。
train.clip_grad_norm：梯度裁剪。对稳定SSL训练至关重要。

Multi-GPU / Multi-Node

多GPU / 多节点

Launch method: Lightning-managed (single

python

process, Lightning spawns workers).

Spec Key	Description	Default
`train.num_gpus`	Number of GPUs	1
`train.gpu_ids`	GPU device indices	[0]
`train.num_nodes`	Number of nodes	1

Strategy:
```
auto
```
(Lightning picks best strategy automatically)
```
sync_batchnorm
```
is always enabled — critical for SSL training with teacher-student framework
Multi-GPU strongly recommended (4-8 GPUs) for meaningful SSL training

Multi-node env vars (set by orchestrator):

WORLD_SIZE

NODE_RANK

MASTER_ADDR

MASTER_PORT

NUM_GPU_PER_NODE

启动方式： Lightning管理（单个

python

进程，Lightning生成工作进程）。

规格键	描述	默认值
`train.num_gpus`	GPU数量	1
`train.gpu_ids`	GPU设备索引	[0]
`train.num_nodes`	节点数量	1

策略：
```
auto
```
（Lightning自动选择最佳策略）
```
sync_batchnorm
```
始终启用——这对基于师生框架的SSL训练至关重要
强烈推荐使用多GPU（4-8块GPU）进行有效的SSL训练

多节点环境变量（由编排器设置）：

WORLD_SIZE

、

NODE_RANK

、

MASTER_ADDR

、

MASTER_PORT

、

NUM_GPU_PER_NODE

。

Hardware

硬件要求

Minimum 4 GPU(s), recommended 8 GPU(s). 40GB+ (A100 recommended) VRAM per GPU. SSL with ViT-Large teacher+student is very memory-intensive. Requires A100 40GB+ GPUs. Multi-GPU strongly recommended.

最少4块GPU，推荐8块GPU。每块GPU需40GB以上显存（推荐A100）。使用ViT-Large师生模型的SSL训练对内存要求极高，需要A100 40GB以上显存的GPU。强烈推荐使用多GPU。

Error Patterns

错误模式

CUDA out of memory: ViT-Large teacher+student with img_size=518 requires 40GB+ GPU memory. Reduce batch_size, img_size, or use smaller ViT variant.

Slow convergence: SSL needs many epochs. Default 10 is for quick testing; production runs typically use 100+ epochs.

CUDA内存不足：使用img_size=518的ViT-Large师生模型需要40GB以上GPU显存。可减小batch_size、img_size，或使用更小的ViT变体。

收敛缓慢：SSL训练需要大量轮次。默认的10轮次仅用于快速测试；生产环境运行通常需要100+轮次。

Spec Param / Parent Model Inference

规格参数 / 父模型推理

Model-specific inference mappings belong in this MD file, not in

config.json

. Generated runners should read this section and apply the mappings with SDK helpers before

create_job()

. This mirrors the old microservices

infer_params.py

flow.

Inference mappings from TAO Core

nvdinov2.config.json

Action	Spec Field	Inference Function	Meaning
distill	`encryption_key`	`key`	encryption key
distill	`model.distill.pretrained_non_distill_pl_model_path`	`parent_model`	model file inferred from the parent job results folder
distill	`results_dir`	`output_dir`	current job results directory
export	`encryption_key`	`key`	encryption key
export	`export.checkpoint`	`parent_model`	model file inferred from the parent job results folder
export	`export.onnx_file`	`create_onnx_file`	output ONNX path
export	`results_dir`	`output_dir`	current job results directory
inference	`encryption_key`	`key`	encryption key
inference	`inference.checkpoint`	`parent_model`	model file inferred from the parent job results folder
inference	`inference.trt_engine`	`parent_model`	model file inferred from the parent job results folder
inference	`results_dir`	`output_dir`	current job results directory
train	`encryption_key`	`key`	encryption key
train	`results_dir`	`output_dir`	current job results directory
train	`train.pretrained_model_path`	`ptm_if_no_resume_model`	PTM when no resume checkpoint exists
train	`train.resume_training_checkpoint_path`	`resume_model`	model file inferred from the current job results folder

For

parent_model

parent_model_folder

, pass the upstream train/export/AutoML child job id as

parent_job_id

. The SDK lists the parent result folder, filters checkpoint artifacts, and returns the selected model file or folder. Do not add these mappings back to

config.json

and do not patch generated runner scripts to guess checkpoint paths.

模型特定的推理映射应放在此MD文件中，而非

config.json

。生成的运行器应读取本节内容，并在

create_job()

前使用SDK助手应用这些映射。这与旧微服务的

infer_params.py

流程一致。

来自TAO Core

nvdinov2.config.json

的推理映射：

操作	规格字段	推理函数	含义
distill	`encryption_key`	`key`	加密密钥
distill	`model.distill.pretrained_non_distill_pl_model_path`	`parent_model`	从父作业结果文件夹推断出的模型文件
distill	`results_dir`	`output_dir`	当前作业结果目录
export	`encryption_key`	`key`	加密密钥
export	`export.checkpoint`	`parent_model`	从父作业结果文件夹推断出的模型文件
export	`export.onnx_file`	`create_onnx_file`	输出ONNX路径
export	`results_dir`	`output_dir`	当前作业结果目录
inference	`encryption_key`	`key`	加密密钥
inference	`inference.checkpoint`	`parent_model`	从父作业结果文件夹推断出的模型文件
inference	`inference.trt_engine`	`parent_model`	从父作业结果文件夹推断出的模型文件
inference	`results_dir`	`output_dir`	当前作业结果目录
train	`encryption_key`	`key`	加密密钥
train	`results_dir`	`output_dir`	当前作业结果目录
train	`train.pretrained_model_path`	`ptm_if_no_resume_model`	无恢复检查点时使用的预训练模型（PTM）
train	`train.resume_training_checkpoint_path`	`resume_model`	从当前作业结果文件夹推断出的模型文件

对于

parent_model

或

parent_model_folder

，传入上游训练/导出/AutoML子作业ID作为

parent_job_id

。SDK会列出父结果文件夹，过滤检查点工件，并返回选定的模型文件或文件夹。请勿将这些映射添加回

config.json

，也不要修改生成的运行器脚本以猜测检查点路径。

Deployment

部署

tao-deploy-nvdinov2 — NvDINOv2 deploy workflow for TensorRT engine generation using TAO Deploy.

tao-deploy-nvdinov2 —— 使用TAO Deploy生成TensorRT引擎的NvDINOv2部署工作流。