monkeypatch.attention.mllama

monkeypatch.attention.mllama

Monkeypatch for Vision Llama for FA2 support

Classes

Name	Description
MllamaTextCrossFlashAttention2	Mllama flash cross-attention module. This module inherits from MllamaTextCrossAttention and
MllamaTextSelfFlashAttention2	Mllama flash self-attention module. This module inherits from MllamaTextSelfAttention and

MllamaTextCrossFlashAttention2

monkeypatch.attention.mllama.MllamaTextCrossFlashAttention2(
    self,
    *args,
    **kwargs,
)

Mllama flash cross-attention module. This module inherits from MllamaTextCrossAttention and implements the forward pass using Flash Attention for improved performance.

MllamaTextSelfFlashAttention2

monkeypatch.attention.mllama.MllamaTextSelfFlashAttention2(
    self,
    config,
    layer_idx,
    *args,
    **kwargs,
)

Mllama flash self-attention module. This module inherits from MllamaTextSelfAttention and implements the forward pass using Flash Attention for improved performance.