Skip to navigation

nemo_automodel.components.models.flux3.types

View as Markdown

Typed component and cache contracts shared by FLUX.3 preprocessors.

Module Contents

Classes

NameDescription
Flux3BucketSignatureLatent shapes used to group compatible image-edit examples.
Flux3CacheDataTensor payload and metadata saved for a text-to-image example.
Flux3CacheMetadataImage metadata supplied by the native preprocessing worker.
Flux3EditRecordJSON index entry for one cached image-edit example.
Flux3ModelsFrozen VAE, normalization buffers, and Qwen3-VL text stack.

API

class nemo_automodel.components.models.flux3.types.Flux3BucketSignature

Bases: typing.TypedDict

Latent shapes used to group compatible image-edit examples.

contexts
list[list[int]]
target
list[int]
class nemo_automodel.components.models.flux3.types.Flux3CacheData()

Bases: Flux3CacheMetadata

Tensor payload and metadata saved for a text-to-image example.

latent
Tensor
model_type
str
prompt_embeds
Tensor
class nemo_automodel.components.models.flux3.types.Flux3CacheMetadata

Bases: typing.TypedDict

Image metadata supplied by the native preprocessing worker.

aspect_ratio
float
bucket_id
int
bucket_resolution
tuple[int, int]
crop_offset
tuple[int, int]
image_path
str
original_resolution
tuple[int, int]
prompt
str
class nemo_automodel.components.models.flux3.types.Flux3EditRecord

Bases: typing.TypedDict

JSON index entry for one cached image-edit example.

cache_file
str
compound_bucket_signature
Flux3BucketSignature
context_token_lengths
list[int]
original_id
str
row_index
int
target_token_length
int
text_token_length
int
class nemo_automodel.components.models.flux3.types.Flux3Models()

Bases: Flux3TextStack

Frozen VAE, normalization buffers, and Qwen3-VL text stack.

bn_mean
Tensor
bn_std
Tensor
vae
AutoencoderKLFlux2