For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Python function

dataclass

`dataclass()`

max.tree.dataclass(cls: type[_T], /) → type[_T]

source

max.tree.dataclass(cls: None = None, /, **kwargs: Any) → Callable[[type[_T]], type[_T]]

Makes a dataclass a tree node.

Usable bare or with dataclasses.dataclass() arguments.

from max import tree

@tree.dataclass
class AttentionInputs:
    layer_idx: TensorValue
    freqs_cis: TensorValue
    indexer: TensorValue | None = None

@tree.dataclass(frozen=True)
class PLEInputs:
    conv_pool: BufferValue
    slot_idx: TensorValue

The generated __tree_flatten__ includes fields that are not None at flatten time, in declaration order. The generated __tree_unflatten__ restores any absent field to None – a field is dropped only when it is None, so a non-None default is not reapplied.

Parameters:

cls – The class to decorate, or None when called with arguments.
kwargs – Passed to dataclasses.dataclass() when cls is not already a dataclass.

Returns:

The decorated class, or a decorator when called with arguments.

dataclass()​

`dataclass()`