monkeypatch.models.llama4.modeling

monkeypatch.models.llama4.modeling

Modified Llama-4 text experts modeling for linearized experts for improved LoRA support

Classes

Name Description
Llama4TextExperts Modified Llama-4 text experts modeling for linearized experts

Llama4TextExperts

monkeypatch.models.llama4.modeling.Llama4TextExperts(config)

Modified Llama-4 text experts modeling for linearized experts

Methods

Name Description
forward Forward method using separate Linear layers for each expert.
forward
monkeypatch.models.llama4.modeling.Llama4TextExperts.forward(hidden_states)

Forward method using separate Linear layers for each expert.

Parameters
Name Type Description Default
hidden_states torch.Tensor (num_experts * batch_size, hidden_size) The input should be organized by expert required
Returns
Name Type Description
torch.Tensor torch.Tensor: (num_experts * batch_size, hidden_size)

Functions

Name Description
patch_llama4_linearized_modeling Patch Llama4TextExperts to use separate Linear layers for each expert.

patch_llama4_linearized_modeling

monkeypatch.models.llama4.modeling.patch_llama4_linearized_modeling()

Patch Llama4TextExperts to use separate Linear layers for each expert.