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

LLM: fix AttributeError of FP16Linear (#10740)

Browse source
This commit is contained in:
binbin Deng 2024-04-11 17:03:56 +08:00 committed by GitHub
parent 1256a2cc4e
commit 70ed9397f9
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 4 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
python/llm/src/ipex_llm/transformers/low_bit_linear.py
View file

@ -729,7 +729,7 @@ class LowBitLinear(nn.Linear):
class FP16Linear(nn.Linear):
def __init__(self, input_features, output_features, bias=True,
mp_group=None, weight_type=1,
mp_group=None, weight_type=1, enable_xetla=False,
optimize_lm_head=False):
super().__init__(input_features, output_features, bias)
self.in_len = input_features
@ -743,6 +743,7 @@ class FP16Linear(nn.Linear):
# weigh_type = 3 means weight has been transposed by esimd method
self.weight_type = 1
self.optimize_lm_head = optimize_lm_head
self.enable_xetla = enable_xetla
def forward(self, x: torch.Tensor):
# only work for GPU
@ -849,7 +850,7 @@ class FP16Linear(nn.Linear):
class BF16Linear(nn.Linear):
def __init__(self, input_features, output_features, bias=True,
mp_group=None, compute_dtype=None,
mp_group=None, compute_dtype=None, enable_xetla=False,
optimize_lm_head=False):
super().__init__(input_features, output_features, bias)
self.in_len = input_features
@ -860,6 +861,7 @@ class BF16Linear(nn.Linear):
self.mp_group = mp_group
self.compute_dtype = compute_dtype
self.optimize_lm_head = optimize_lm_head
self.enable_xetla = enable_xetla
def forward(self, x: torch.Tensor):
if self.optimize_lm_head: