fix awq_dequantize import (#5669)
This commit is contained in:
@@ -3,7 +3,6 @@ import logging
|
||||
from typing import Any, Dict, List, Optional
|
||||
|
||||
import torch
|
||||
from sgl_kernel import awq_dequantize
|
||||
|
||||
from sglang.srt.layers.linear import (
|
||||
LinearBase,
|
||||
@@ -12,6 +11,11 @@ from sglang.srt.layers.linear import (
|
||||
)
|
||||
from sglang.srt.layers.parameter import GroupQuantScaleParameter, PackedvLLMParameter
|
||||
from sglang.srt.layers.quantization.base_config import QuantizationConfig
|
||||
from sglang.srt.utils import is_cuda
|
||||
|
||||
_is_cuda = is_cuda()
|
||||
if _is_cuda:
|
||||
from sgl_kernel import awq_dequantize
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user