ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

small fix (#12634)

Browse source
This commit is contained in:
Yishuo Wang 2024-12-30 17:14:25 +08:00 committed by GitHub
parent 2d08155513
commit f289f68d57
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 1 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
python/llm/src/ipex_llm/transformers/low_bit_linear.py
View file

@ -644,7 +644,6 @@ class LowBitLinear(nn.Linear):
if x0.device.type == "xpu": if x0.device.type == "xpu":
# GPU logic # GPU logic
try: try:
import intel_extension_for_pytorch
import xe_linear import xe_linear
from ipex_llm.transformers.models.utils import use_xmx from ipex_llm.transformers.models.utils import use_xmx
except ModuleNotFoundError: except ModuleNotFoundError:

3
python/llm/src/ipex_llm/transformers/models/utils.py
View file

@ -346,8 +346,7 @@ def use_decoding_fast_path(proj,
def use_xmx(x: torch.Tensor, qtype: int): def use_xmx(x: torch.Tensor, qtype: int):
device = get_xpu_device_type(x) device = get_xpu_device_type(x)
return ( return (
os.environ.get("BIGDL_LLM_XMX_DISABLED", "0") != "1" device in ["arc", "flex", "pvc"]
and device in ["arc", "flex", "pvc"]
and qtype in [SYM_INT4, SYM_INT8, FP8E4, FP8E5] and qtype in [SYM_INT4, SYM_INT8, FP8E4, FP8E5]
and ( and (
(device == "pvc" and 1 < x.size(0) <= 16) (device == "pvc" and 1 < x.size(0) <= 16)