[Refactor] Consolidate SupportsEagle (#36063)
Signed-off-by: Benjamin Chislett <bchislett@nvidia.com>
This commit is contained in:
@@ -89,6 +89,7 @@ from vllm.v1.attention.backends.registry import AttentionBackendEnum
|
||||
|
||||
from .interfaces import (
|
||||
MultiModalEmbeddings,
|
||||
SupportsEagle,
|
||||
SupportsEagle3,
|
||||
SupportsLoRA,
|
||||
SupportsMRoPE,
|
||||
@@ -1000,6 +1001,7 @@ class Qwen2_5_VLForConditionalGeneration(
|
||||
SupportsLoRA,
|
||||
SupportsPP,
|
||||
SupportsQuant,
|
||||
SupportsEagle,
|
||||
SupportsEagle3,
|
||||
SupportsMultiModalPruning,
|
||||
SupportsMRoPE,
|
||||
@@ -1143,13 +1145,6 @@ class Qwen2_5_VLForConditionalGeneration(
|
||||
self.language_model.make_empty_intermediate_tensors
|
||||
)
|
||||
|
||||
def set_aux_hidden_state_layers(self, layers: tuple[int, ...]) -> None:
|
||||
self.language_model.model.aux_hidden_state_layers = layers
|
||||
|
||||
def get_eagle3_aux_hidden_state_layers(self) -> tuple[int, ...]:
|
||||
num_layers = len(self.language_model.model.layers)
|
||||
return (2, num_layers // 2, num_layers - 3)
|
||||
|
||||
def _parse_and_validate_image_input(
|
||||
self, **kwargs: object
|
||||
) -> Qwen2_5_VLImageInputs | None:
|
||||
|
||||
Reference in New Issue
Block a user