Is this page helpful?

`core.transformer.mla_layer_config`#

Module Contents#

Classes#

MLALayerConfig

Configuration for a Multi-Latent Attention layer in a hybrid stack.

API#

class core.transformer.mla_layer_config.MLALayerConfig#

Bases: megatron.core.transformer.transformer_config.MLATransformerConfig

Configuration for a Multi-Latent Attention layer in a hybrid stack.

Due to backwards-compatibility, this config’s arguments are defined in MLATransformerConfig.