[Bugfix] Lazily import the humming quantization backend (#44921)
Signed-off-by: mgoin <mgoin64@gmail.com> Co-authored-by: Claude <noreply@anthropic.com>
This commit is contained in:
@@ -5,8 +5,6 @@ from typing import Any
|
||||
|
||||
import regex as re
|
||||
import torch
|
||||
from humming.layer import HummingInputSchema, HummingMethod
|
||||
from humming.schema import BaseWeightSchema
|
||||
|
||||
from vllm import envs
|
||||
from vllm.model_executor.layers.fused_moe import RoutedExperts
|
||||
@@ -16,6 +14,7 @@ from vllm.model_executor.layers.fused_moe.config import (
|
||||
)
|
||||
from vllm.model_executor.layers.linear import LinearBase
|
||||
from vllm.model_executor.layers.quantization.utils.quant_utils import GroupShape
|
||||
from vllm.utils.humming import BaseWeightSchema, HummingInputSchema, HummingMethod
|
||||
|
||||
|
||||
def humming_is_layer_skipped(config: dict[str, Any], prefix: str):
|
||||
|
||||
Reference in New Issue
Block a user