CustomizationTarget#

class nemo_microservices.types.CustomizationTarget(*args: Any, **kwargs: Any)

Bases: BaseModel

model_path: str

Path to the model checkpoints to use for training.

Absolute path or local path from the models cache

num_parameters: int

Number of parameters used for training the model

precision: Literal['int8', 'bf16', 'fp16', 'fp32', 'fp8-mixed', 'bf16-mixed']

Type of model precision.

## Values

  • “int8” - 8-bit integer precision

  • “bf16” - Brain floating point precision

  • “fp16” - 16-bit floating point precision

  • “fp32” - 32-bit floating point precision

  • “fp8-mixed” - Mixed 8-bit floating point precision available on Hopper and later architectures.

  • “bf16-mixed” - Mixed Brain floating point precision

id: str | None = None

The ID of the entity.

With the exception of namespaces, this is always a semantically-prefixed base58-encoded uuid4 [<prefix>-base58(uuid4())].

base_model: str | None = None

Default to being the same as the the configuration entry name, maps to the name in NIM

created_at: datetime | None = None

Timestamp for when the entity was created.

description: str | None = None

The description of the entity.

enabled: bool | None = None

Enable the model for training jobs

hf_endpoint: str | None = None

Configure HuggingFace Hub base URL.

Defaults to NeMo Data Store. Set value as “https://huggingface.co” to download model_uri from HuggingFace.

model_type: Literal['nemo', 'hf', 'hf-lora'] | None = None

The detected checkpoint type of the uploaded target.

nemo model checkpoints have these key files: context/nemo_tokenizer/config.json, context/model.yaml, and weights/metadata.json

hf model checkpoints have these key files: config.json, tokenizer.json or tokenizer_config.json, and either model*.safetensors (preferred) or pytorch_model*.bin

hf-lora model checkpoints only contain the LoRA adapter for a HF model, they have these key files: adapter_config.json adapter_model.safetensors

model_uri: str | None = None

The URI of the model to download to the model cache at the model_path directory.

To download from NGC, specify ngc://org/optional-team/model-name:version. To download from Nemo Data Store, specify hf://namespace/model-name@checkpoint-name

name: str | None = None

The name of the entity.

Must be unique inside the namespace. If not specified, it will be the same as the automatically generated id.

namespace: str | None = None

The namespace of the entity.

This can be missing for namespace entities or in deployments that don’t use namespaces.

ownership: Ownership | None = None

Information about ownership of an entity.

If the entity is a namespace, the access_policies will typically apply to all entities inside the namespace.

project: str | None = None

The URN of the project associated with this entity.

status: Literal['created', 'pending', 'downloading', 'failed', 'ready', 'cancelled', 'unknown', 'deleted', 'deleting', 'delete_failed'] | None = None

Normalized statuses for targets.

  • CREATED: The target is created, but not yet scheduled.

  • PENDING: The target is waiting for resource allocation.

  • DOWNLOADING: The target is downloading.

  • FAILED: The target failed to execute and terminated.

  • READY: The target is ready to be used.

  • CANCELLED: The target download was cancelled.

  • UNKNOWN: The target status is unknown.

  • DELETED: The target is deleted.

  • DELETING: The target is currently being deleted.

  • DELETE_FAILED: Failed to delete the target.

tokenizer: object | None = None

Overrides for the model tokenizer

updated_at: datetime | None = None

Timestamp for when the entity was last updated.