apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-11-03 02:53:26 +08:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

[Sync] Update to latest code (#2679)

Browse Source 
* [Sync] Update to latest code

* Add new code files

* Add new code files

* update code

* Try to fix build.sh

* Try to fix build.sh

* Update code

* Update requirements.txt

* Update code

---------

Co-authored-by: Jiang-Jia-Jun <jiangjiajun@baidu.com>
This commit is contained in:
Jiang-Jia-Jun
2025-07-03 15:43:53 +08:00
committed by GitHub
parent d222248d00
commit 05c670e593
95 changed files with 9916 additions and 1312 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
fastdeploy/model_executor/layers/backends/xpu/quantization/weight_only.py
View File

@@ -41,16 +41,12 @@ class XPUWeightOnlyLinearMethod(WeightOnlyLinearMethod):
"""
Create weights for linear layer on XPU
"""
# The scale shape should be equal to the output dim of weight using Per-Channel Quantization.
linear_weight_scale_shape = [layer.linear_weight_shape[1]]
layer.linear_weight_shape.reverse()
if self.quant_config.name() == "weight_only_int4":
layer.linear_weight_shape[0] //= 2
layer.weight_dtype = "int8"
linear_weight_scale_shape = [layer.embed_dim]
if hasattr(layer, "linear_weight_shape"):
if isinstance(layer.linear_weight_shape, list):
layer_weight_shape = layer.linear_weight_shape
linear_weight_scale_shape = layer_weight_shape[:1]
layer.linear_weight_scale = layer.create_parameter(
shape=linear_weight_scale_shape,
dtype="float32",