[MoE Refactor] FusedMoE/MoERunner inversion refactor (#41184)
Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
+4
-4
@@ -33,7 +33,7 @@ from vllm.lora.layers import (
|
||||
RowParallelLinearWithShardedLoRA,
|
||||
VocabParallelEmbeddingWithLoRA,
|
||||
)
|
||||
from vllm.model_executor.layers.fused_moe import FusedMoE
|
||||
from vllm.model_executor.layers.fused_moe import MoERunner
|
||||
from vllm.model_executor.layers.linear import LinearBase
|
||||
from vllm.model_executor.utils import get_moe_expert_mapping, get_packed_modules_mapping
|
||||
from vllm.transformers_utils.repo_utils import hf_api
|
||||
@@ -96,8 +96,8 @@ _all_lora_classes: tuple[type[BaseLayerWithLoRA], ...] = (
|
||||
|
||||
|
||||
def is_moe_model(model: nn.Module) -> bool:
|
||||
"""Checks if the model contains FusedMoE layers and warns the user."""
|
||||
if any(isinstance(module, FusedMoE) for module in model.modules()):
|
||||
"""Checks if the model contains MoERunner layers and warns the user."""
|
||||
if any(isinstance(module, MoERunner) for module in model.modules()):
|
||||
logger.info_once("MoE model detected. Using fused MoE LoRA implementation.")
|
||||
return True
|
||||
return False
|
||||
@@ -223,7 +223,7 @@ def get_supported_lora_modules(model: nn.Module) -> list[str]:
|
||||
if isinstance(module, (LinearBase,)):
|
||||
supported_lora_modules.add(name.split(".")[-1])
|
||||
|
||||
if isinstance(module, (FusedMoE,)):
|
||||
if isinstance(module, (MoERunner,)):
|
||||
supported_lora_modules.add(name.split(".")[-1])
|
||||
|
||||
return list(supported_lora_modules)
|
||||
|
||||
Reference in New Issue
Block a user