thunder.distributed.FSDPBucketingStrategy

class thunder.distributed.FSDPBucketingStrategy(value)[source]

Bases: Enum

Specify how we group parameters into a bucket for collective communication in fsdp.

NONE[source]

Disables bucketing.

LAYER[source]

Creates buckets per layer such as torch.nn.Linear and torch.nn.LayerNorm.

BLOCK[source]

Creates buckets per block such as Transformer block.

__init__()[source]

Attributes

NONE

LAYER

BLOCK

Was this page helpful?

What went wrong?

↑↓ Navigate↵ Open⌘↵ New tabesc Close