Skip to content

vllm.lora.layers.base

Classes:

BaseLayerWithLoRA

Bases: Module

Methods:

Source code in vllm/lora/layers/base.py 
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
class BaseLayerWithLoRA(nn.Module):
    @overload
    def slice_lora_a(
        self, lora_a: list[torch.Tensor | None]
    ) -> list[torch.Tensor | None]: ...
    @overload
    def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor: ...
    def slice_lora_a(
        self, lora_a: torch.Tensor | list[torch.Tensor | None]
    ) -> torch.Tensor | list[torch.Tensor | None]:
        """Slice lora a if splitting for tensor parallelism."""
        ...

    @overload
    def slice_lora_b(
        self, lora_b: list[torch.Tensor | None]
    ) -> list[torch.Tensor | None]: ...
    @overload
    def slice_lora_b(self, lora_b: torch.Tensor) -> torch.Tensor: ...
    def slice_lora_b(
        self, lora_b: torch.Tensor | list[torch.Tensor | None]
    ) -> torch.Tensor | list[torch.Tensor | None]:
        """Slice lora b if splitting with tensor parallelism."""
        ...

    def create_lora_weights(
        self,
        max_loras: int,
        lora_config: LoRAConfig,
        model_config: PretrainedConfig | None = None,
    ) -> None:
        """Initializes lora matrices."""
        ...

    def reset_lora(self, index: int):
        """Resets the lora weights at index back to 0."""
        ...

    def set_lora(
        self,
        index: int,
        lora_a: torch.Tensor | list[torch.Tensor],
        lora_b: torch.Tensor | list[torch.Tensor],
    ):
        """Overwrites lora tensors at index."""
        ...

    def set_mapping(
        self,
        punica_wrapper,
    ):
        self.punica_wrapper: PunicaWrapperBase = punica_wrapper

    @classmethod
    def can_replace_layer(
        cls,
        source_layer: nn.Module,
        lora_config: LoRAConfig,
        packed_modules_list: list,
        model_config: PretrainedConfig | None = None,
    ) -> bool:
        """Returns True if the layer can be replaced by this LoRA layer."""
        raise NotImplementedError

can_replace_layer(source_layer, lora_config, packed_modules_list, model_config=None) classmethod

Returns True if the layer can be replaced by this LoRA layer.

Source code in vllm/lora/layers/base.py 
69
70
71
72
73
74
75
76
77
78
@classmethod
def can_replace_layer(
    cls,
    source_layer: nn.Module,
    lora_config: LoRAConfig,
    packed_modules_list: list,
    model_config: PretrainedConfig | None = None,
) -> bool:
    """Returns True if the layer can be replaced by this LoRA layer."""
    raise NotImplementedError

create_lora_weights(max_loras, lora_config, model_config=None)

Initializes lora matrices.

Source code in vllm/lora/layers/base.py 
41
42
43
44
45
46
47
48
def create_lora_weights(
    self,
    max_loras: int,
    lora_config: LoRAConfig,
    model_config: PretrainedConfig | None = None,
) -> None:
    """Initializes lora matrices."""
    ...

reset_lora(index)

Resets the lora weights at index back to 0.

Source code in vllm/lora/layers/base.py 
50
51
52
def reset_lora(self, index: int):
    """Resets the lora weights at index back to 0."""
    ...

set_lora(index, lora_a, lora_b)

Overwrites lora tensors at index.

Source code in vllm/lora/layers/base.py 
54
55
56
57
58
59
60
61
def set_lora(
    self,
    index: int,
    lora_a: torch.Tensor | list[torch.Tensor],
    lora_b: torch.Tensor | list[torch.Tensor],
):
    """Overwrites lora tensors at index."""
    ...

slice_lora_a(lora_a) 

slice_lora_a(
    lora_a: list[torch.Tensor | None],
) -> list[torch.Tensor | None]

slice_lora_a(lora_a: torch.Tensor) -> torch.Tensor

Slice lora a if splitting for tensor parallelism.

Source code in vllm/lora/layers/base.py 
23
24
25
26
27
def slice_lora_a(
    self, lora_a: torch.Tensor | list[torch.Tensor | None]
) -> torch.Tensor | list[torch.Tensor | None]:
    """Slice lora a if splitting for tensor parallelism."""
    ...

slice_lora_b(lora_b) 

slice_lora_b(
    lora_b: list[torch.Tensor | None],
) -> list[torch.Tensor | None]

slice_lora_b(lora_b: torch.Tensor) -> torch.Tensor

Slice lora b if splitting with tensor parallelism.

Source code in vllm/lora/layers/base.py 
35
36
37
38
39
def slice_lora_b(
    self, lora_b: torch.Tensor | list[torch.Tensor | None]
) -> torch.Tensor | list[torch.Tensor | None]:
    """Slice lora b if splitting with tensor parallelism."""
    ...