Back to Details

tao-train-mask2former

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Mask2Former

Mask2Former

Mask2Former for universal image segmentation (panoptic, instance, and semantic). Transformer-based with masked attention for high-quality segmentation results.
Set model.backbone.pretrained_weights for Swin backbone weights.
For TAO Deploy TensorRT actions (
gen_trt_engine
, TensorRT 
evaluate
, and TensorRT 
inference
), read 
references/tao-deploy-mask2former.md
first. Deploy spec templates live in this skill's 
references/
folder with the 
spec_template_deploy_*.yaml
prefix.
Mask2Former用于通用图像分割（全景分割、实例分割和语义分割）。这是基于Transformer的模型，采用掩码注意力机制以实现高质量的分割结果。
设置model.backbone.pretrained_weights以加载Swin骨干网络权重。
对于TAO Deploy TensorRT操作（
gen_trt_engine
、TensorRT 
evaluate
和TensorRT 
inference
），请先阅读
references/tao-deploy-mask2former.md
。部署配置模板存放在本技能的
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
列出了可用的操作。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
；维护人员在打包技能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。处理任何训练阶段的请求前，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，并使用该模型的
skill_dir
。保留数据集、配置、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）仍在本模型技能中执行。每次运行的
automl_policy
覆盖配置不会改变模型元数据。

Training Requirements

训练要求

  • Dataset type: segmentation
  • Formats: coco_panoptic, coco
  • Monitoring metric: mIoU
  • 数据集类型： 分割数据集
  • 格式： coco_panoptic、coco
  • 监控指标： mIoU

Per-Action Dataset Requirements

各操作的数据集要求

ActionSpec KeySourceFilesList?
evaluatedataset.train.img_dirtrain_datasetsimages.tar.gzNo
evaluatedataset.label_maptrain_datasetscoco_panoptic: label_map_panoptic.json; *: label_map.jsonNo
evaluatedataset.train.instance_jsontrain_datasetsannotations.jsonNo
evaluatedataset.train.panoptic_jsontrain_datasetsannotations_panoptic.jsonNo
evaluatedataset.train.panoptic_dirtrain_datasetsimages_panoptic.tar.gzNo
evaluatedataset.val.img_direval_datasetimages.tar.gzNo
evaluatedataset.val.instance_jsoneval_datasetannotations.jsonNo
evaluatedataset.val.panoptic_jsoneval_datasetannotations_panoptic.jsonNo
evaluatedataset.val.panoptic_direval_datasetimages_panoptic.tar.gzNo
evaluatedataset.test.img_direval_datasetimages.tar.gzNo
inferencedataset.train.img_dirtrain_datasetsimages.tar.gzNo
inferencedataset.label_maptrain_datasetscoco_panoptic: label_map_panoptic.json; *: label_map.jsonNo
inferencedataset.train.instance_jsontrain_datasetsannotations.jsonNo
inferencedataset.train.panoptic_jsontrain_datasetsannotations_panoptic.jsonNo
inferencedataset.train.panoptic_dirtrain_datasetsimages_panoptic.tar.gzNo
inferencedataset.val.img_direval_datasetimages.tar.gzNo
inferencedataset.val.instance_jsoneval_datasetannotations.jsonNo
inferencedataset.val.panoptic_jsoneval_datasetannotations_panoptic.jsonNo
inferencedataset.val.panoptic_direval_datasetimages_panoptic.tar.gzNo
inferencedataset.test.img_direval_datasetimages.tar.gzNo
quantizedataset.train.img_dirtrain_datasetsimages.tar.gzNo
quantizedataset.label_maptrain_datasetscoco_panoptic: label_map_panoptic.json; *: label_map.jsonNo
quantizedataset.train.instance_jsontrain_datasetsannotations.jsonNo
quantizedataset.train.panoptic_jsontrain_datasetsannotations_panoptic.jsonNo
quantizedataset.train.panoptic_dirtrain_datasetsimages_panoptic.tar.gzNo
quantizedataset.val.img_direval_datasetimages.tar.gzNo
quantizedataset.val.instance_jsoneval_datasetannotations.jsonNo
quantizedataset.val.panoptic_jsoneval_datasetannotations_panoptic.jsonNo
quantizedataset.val.panoptic_direval_datasetimages_panoptic.tar.gzNo
quantizedataset.test.img_direval_datasetimages.tar.gzNo
quantizedataset.quant_calibration_dataset.images_dirtrain_datasetsimages.tar.gzNo
traindataset.train.img_dirtrain_datasetsimages.tar.gzNo
traindataset.label_maptrain_datasetscoco_panoptic: label_map_panoptic.json; *: label_map.jsonNo
traindataset.train.instance_jsontrain_datasetsannotations.jsonNo
traindataset.train.panoptic_jsontrain_datasetsannotations_panoptic.jsonNo
traindataset.train.panoptic_dirtrain_datasetsimages_panoptic.tar.gzNo
traindataset.val.img_direval_datasetimages.tar.gzNo
traindataset.val.instance_jsoneval_datasetannotations.jsonNo
traindataset.val.panoptic_jsoneval_datasetannotations_panoptic.jsonNo
traindataset.val.panoptic_direval_datasetimages_panoptic.tar.gzNo
traindataset.test.img_direval_datasetimages.tar.gzNo
操作配置键来源文件是否为列表？
evaluatedataset.train.img_dirtrain_datasetsimages.tar.gz否
evaluatedataset.label_maptrain_datasetscoco_panoptic: label_map_panoptic.json; *: label_map.json否
evaluatedataset.train.instance_jsontrain_datasetsannotations.json否
evaluatedataset.train.panoptic_jsontrain_datasetsannotations_panoptic.json否
evaluatedataset.train.panoptic_dirtrain_datasetsimages_panoptic.tar.gz否
evaluatedataset.val.img_direval_datasetimages.tar.gz否
evaluatedataset.val.instance_jsoneval_datasetannotations.json否
evaluatedataset.val.panoptic_jsoneval_datasetannotations_panoptic.json否
evaluatedataset.val.panoptic_direval_datasetimages_panoptic.tar.gz否
evaluatedataset.test.img_direval_datasetimages.tar.gz否
inferencedataset.train.img_dirtrain_datasetsimages.tar.gz否
inferencedataset.label_maptrain_datasetscoco_panoptic: label_map_panoptic.json; *: label_map.json否
inferencedataset.train.instance_jsontrain_datasetsannotations.json否
inferencedataset.train.panoptic_jsontrain_datasetsannotations_panoptic.json否
inferencedataset.train.panoptic_dirtrain_datasetsimages_panoptic.tar.gz否
inferencedataset.val.img_direval_datasetimages.tar.gz否
inferencedataset.val.instance_jsoneval_datasetannotations.json否
inferencedataset.val.panoptic_jsoneval_datasetannotations_panoptic.json否
inferencedataset.val.panoptic_direval_datasetimages_panoptic.tar.gz否
inferencedataset.test.img_direval_datasetimages.tar.gz否
quantizedataset.train.img_dirtrain_datasetsimages.tar.gz否
quantizedataset.label_maptrain_datasetscoco_panoptic: label_map_panoptic.json; *: label_map.json否
quantizedataset.train.instance_jsontrain_datasetsannotations.json否
quantizedataset.train.panoptic_jsontrain_datasetsannotations_panoptic.json否
quantizedataset.train.panoptic_dirtrain_datasetsimages_panoptic.tar.gz否
quantizedataset.val.img_direval_datasetimages.tar.gz否
quantizedataset.val.instance_jsoneval_datasetannotations.json否
quantizedataset.val.panoptic_jsoneval_datasetannotations_panoptic.json否
quantizedataset.val.panoptic_direval_datasetimages_panoptic.tar.gz否
quantizedataset.test.img_direval_datasetimages.tar.gz否
quantizedataset.quant_calibration_dataset.images_dirtrain_datasetsimages.tar.gz否
traindataset.train.img_dirtrain_datasetsimages.tar.gz否
traindataset.label_maptrain_datasetscoco_panoptic: label_map_panoptic.json; *: label_map.json否
traindataset.train.instance_jsontrain_datasetsannotations.json否
traindataset.train.panoptic_jsontrain_datasetsannotations_panoptic.json否
traindataset.train.panoptic_dirtrain_datasetsimages_panoptic.tar.gz否
traindataset.val.img_direval_datasetimages.tar.gz否
traindataset.val.instance_jsoneval_datasetannotations.json否
traindataset.val.panoptic_jsoneval_datasetannotations_panoptic.json否
traindataset.val.panoptic_direval_datasetimages_panoptic.tar.gz否
traindataset.test.img_direval_datasetimages.tar.gz否

Typical Spec Overrides

典型配置覆盖

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,
    "train.validation_interval": 10,
    "model.sem_seg_head.num_classes": 90,
    "dataset.contiguous_id": True,
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": {"coco_panoptic": f"{S3_TRAIN}/label_map_panoptic.json; *: label_map.json"},
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
}
evaluate (mandatory data sources):
python
{
    "model.sem_seg_head.num_classes": 90,
    "dataset.contiguous_id": True,
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": {"coco_panoptic": f"{S3_TRAIN}/label_map_panoptic.json; *: label_map.json"},
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
}
export:
python
{
    "model.sem_seg_head.num_classes": 90,
}
inference (mandatory data sources):
python
{
    "model.sem_seg_head.num_classes": 90,
    "dataset.contiguous_id": True,
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": {"coco_panoptic": f"{S3_TRAIN}/label_map_panoptic.json; *: label_map.json"},
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
}
quantize (mandatory data sources):
python
{
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": {"coco_panoptic": f"{S3_TRAIN}/label_map_panoptic.json; *: label_map.json"},
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.quant_calibration_dataset.images_dir": f"{S3_TRAIN}/images.tar.gz",
}
数据源覆盖对每个操作都是必需的——智能体必须根据上述各操作的数据集要求表构建数据源路径，并将其包含在
spec_overrides
中。
python
S3_TRAIN = "s3://bucket/data/train"
S3_EVAL = "s3://bucket/data/eval"
train（必需数据源）：
python
{
    "train.num_gpus": 1,
    "train.num_epochs": 10,
    "train.checkpoint_interval": 10,
    "train.validation_interval": 10,
    "model.sem_seg_head.num_classes": 90,
    "dataset.contiguous_id": True,
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": {"coco_panoptic": f"{S3_TRAIN}/label_map_panoptic.json; *: label_map.json"},
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
}
evaluate（必需数据源）：
python
{
    "model.sem_seg_head.num_classes": 90,
    "dataset.contiguous_id": True,
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": {"coco_panoptic": f"{S3_TRAIN}/label_map_panoptic.json; *: label_map.json"},
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
}
export：
python
{
    "model.sem_seg_head.num_classes": 90,
}
inference（必需数据源）：
python
{
    "model.sem_seg_head.num_classes": 90,
    "dataset.contiguous_id": True,
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": {"coco_panoptic": f"{S3_TRAIN}/label_map_panoptic.json; *: label_map.json"},
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
}
quantize（必需数据源）：
python
{
    "dataset.train.img_dir": f"{S3_TRAIN}/images.tar.gz",
    "dataset.label_map": {"coco_panoptic": f"{S3_TRAIN}/label_map_panoptic.json; *: label_map.json"},
    "dataset.train.instance_json": f"{S3_TRAIN}/annotations.json",
    "dataset.train.panoptic_json": f"{S3_TRAIN}/annotations_panoptic.json",
    "dataset.train.panoptic_dir": f"{S3_TRAIN}/images_panoptic.tar.gz",
    "dataset.val.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.val.instance_json": f"{S3_EVAL}/annotations.json",
    "dataset.val.panoptic_json": f"{S3_EVAL}/annotations_panoptic.json",
    "dataset.val.panoptic_dir": f"{S3_EVAL}/images_panoptic.tar.gz",
    "dataset.test.img_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.quant_calibration_dataset.images_dir": f"{S3_TRAIN}/images.tar.gz",
}

Eval Dataset

评估数据集

Optional. Val data sources are part of the dataset config alongside train.
可选。验证数据源与训练数据源一同包含在数据集配置中。

Important Parameters

重要参数

  • model.sem_seg_head.num_classes: Number of segmentation classes. Default 200. Must match your annotation categories.
  • model.backbone.swin.type: Swin Transformer variant. Default tiny. Options include tiny, small, base, large.
  • model.mode: Segmentation mode. Default panoptic. Options: panoptic, instance, semantic.
  • train.optim.lr: Learning rate. Default 2e-4 (AdamW).
  • dataset.train.batch_size: Per-GPU batch size. Default 1. Mask2Former is memory-intensive due to per-pixel predictions.
  • model.sem_seg_head.num_classes：分割类别数量。默认值为200。必须与你的标注类别数量匹配。
  • model.backbone.swin.type：Swin Transformer变体。默认值为tiny。选项包括tiny、small、base、large。
  • model.mode：分割模式。默认值为panoptic。选项：panoptic、instance、semantic。
  • train.optim.lr：学习率。默认值为2e-4（AdamW优化器）。
  • dataset.train.batch_size：单GPU批次大小。默认值为1。由于Mask2Former需要逐像素预测，因此对内存要求较高。

Multi-GPU / Multi-Node

多GPU/多节点

Launch method: Lightning-managed (single 
python
process, Lightning spawns workers).
Spec KeyDescriptionDefault
train.num_gpus
Number of GPUs1
train.gpu_ids
GPU device indices[0]
train.num_nodes
Number of nodes1
train.distributed_strategy
ddp
or 
fsdp
ddp
  • Same DDP/FSDP behavior as DINO (activation checkpoint aware)
  • FAN backbones auto-enable 
    sync_batchnorm
  • fsdp
    forces FP16
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）。
配置键描述默认值
train.num_gpus
GPU数量1
train.gpu_ids
GPU设备索引[0]
train.num_nodes
节点数量1
train.distributed_strategy
ddp
或
fsdp
ddp
  • 与DINO相同的DDP/FSDP行为（支持激活检查点）
  • FAN骨干网络自动启用
    sync_batchnorm
  • fsdp
    强制使用FP16
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
。

Export / TRT Defaults

导出/TRT默认值

  • TRT data types: FP32, FP16 only — INT8 is NOT supported
Full TAO Deploy reference: tao-deploy-mask2former.
  • TRT数据类型：仅支持FP32、FP16 —— 不支持INT8
完整的TAO Deploy参考：tao-deploy-mask2former。

Hardware

硬件要求

Minimum 1 GPU(s), recommended 4 GPU(s). 24GB+ (A100 recommended) VRAM per GPU. Mask2Former is memory-heavy. batch_size=1 is the default for good reason. Multi-GPU recommended for reasonable training speed.
最少1块GPU，推荐4块GPU。每块GPU需24GB及以上显存（推荐A100）。Mask2Former内存占用较高。默认批次大小设为1是有原因的。推荐使用多GPU以保证合理的训练速度。

Error Patterns

错误模式

CUDA out of memory: batch_size is already 1 by default. Reduce image resolution in augmentation config or use a smaller Swin variant.
Panoptic vs instance format mismatch: Ensure you provide the correct annotation format matching model.mode setting.
CUDA内存不足：默认批次大小已经是1。可在增强配置中降低图像分辨率，或使用更小的Swin变体。
全景分割与实例分割格式不匹配：确保你提供的标注格式与model.mode设置一致。

Spec Param / Parent Model Inference

配置参数/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 
mask2former.config.json
:
ActionSpec FieldInference FunctionMeaning
evaluate
encryption_key
key
encryption key
evaluate
evaluate.checkpoint
parent_model
model file inferred from the parent job results folder
evaluate
evaluate.trt_engine
parent_model
model file inferred from the parent job results folder
evaluate
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
gen_trt_engine
encryption_key
key
encryption key
gen_trt_engine
gen_trt_engine.onnx_file
parent_model
model file inferred from the parent job results folder
gen_trt_engine
gen_trt_engine.trt_engine
create_engine_file
output TensorRT engine path
gen_trt_engine
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
quantize
encryption_key
key
encryption key
quantize
quantize.model_path
parent_model
model file inferred from the parent job results folder
quantize
results_dir
output_dir
current job results directory
train
encryption_key
key
encryption key
train
model.backbone.pretrained_weights
{'link': 'https://github.com/SwinTransformer/storage/releases/download/v1.0.8/swin_tiny_patch4_window7_224_22k.pth', 'destination_path': '/ptm/mask2former/swin_tiny_patch4_window7_224_22k/swin_tiny_patch4_window7_224_22k.pth'}
{'link': 'https://github.com/SwinTransformer/storage/releases/download/v1.0.8/swin_tiny_patch4_window7_224_22k.pth', 'destination_path': '/ptm/mask2former/swin_tiny_patch4_window7_224_22k/swin_tiny_patch4_window7_224_22k.pth'}
train
results_dir
output_dir
current job results directory
train
train.resume_training_checkpoint_path
resume_model
model file inferred from the current job results folder
For 
parent_model
or 
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本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助手应用映射。这与旧微服务的
infer_params.py
流程一致。
来自TAO Core 
mask2former.config.json
的推理映射：
操作配置字段推理函数含义
evaluate
encryption_key
key
加密密钥
evaluate
evaluate.checkpoint
parent_model
从父作业结果文件夹推断出的模型文件
evaluate
evaluate.trt_engine
parent_model
从父作业结果文件夹推断出的模型文件
evaluate
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
gen_trt_engine
encryption_key
key
加密密钥
gen_trt_engine
gen_trt_engine.onnx_file
parent_model
从父作业结果文件夹推断出的模型文件
gen_trt_engine
gen_trt_engine.trt_engine
create_engine_file
输出TensorRT引擎路径
gen_trt_engine
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
quantize
encryption_key
key
加密密钥
quantize
quantize.model_path
parent_model
从父作业结果文件夹推断出的模型文件
quantize
results_dir
output_dir
当前作业结果目录
train
encryption_key
key
加密密钥
train
model.backbone.pretrained_weights
{'link': 'https://github.com/SwinTransformer/storage/releases/download/v1.0.8/swin_tiny_patch4_window7_224_22k.pth', 'destination_path': '/ptm/mask2former/swin_tiny_patch4_window7_224_22k/swin_tiny_patch4_window7_224_22k.pth'}
{'link': 'https://github.com/SwinTransformer/storage/releases/download/v1.0.8/swin_tiny_patch4_window7_224_22k.pth', 'destination_path': '/ptm/mask2former/swin_tiny_patch4_window7_224_22k/swin_tiny_patch4_window7_224_22k.pth'}
train
results_dir
output_dir
当前作业结果目录
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，筛选检查点工件，并返回选定的模型文件或文件夹。不要将这些映射添加回
config.json
，也不要修改生成的运行器脚本以猜测检查点路径。