vllm.model_executor.layers.attention_layer_base
Base class for attention-like layers.
AttentionLayerBase ¶
Bases: ABC
Base class for attention-like layers (Attention, Mamba, etc.) that support the v1 engine.
This provides a common interface for getting attention backends from different layer types.
Source code in vllm/model_executor/layers/attention_layer_base.py
get_attn_backend abstractmethod
¶
get_attn_backend() -> type[AttentionBackend]