ipex 2.7.10+xpu run CosyVoice2-0.5B on xpu is much slower than cpu

[wenkema]

Describe the issue

Describe the issue
Recently, refer to ipex example, we have made CosyVoice2-0.5B (same like a Qwen2-0.5B model) run on xpu.
this is my change to use xpu
but the performance is poor compared to cpu.
Our environment is as follows:
OS: ubuntu 24.04
kernel：6.8.0-59-generic
CPU:Intel(R) Xeon(R) w5-3425
GPU: Intel ARC A770
Setup up steps:
1 install Ubuntu 24.04.
2 refer to Client GPU driver install A770 driver
3 refer to ipex install guide to install torch 2.7.0 and ipex 2.7.10+xpu
4 install conda and run cosyvoice test on xpu

this is perfermance data:

Time(s)	Llm model	Flow model	Hift model	Full inference
cpu	8.75	4.70	0.43	14.18
xpu	66.91	5.62	1.32	74.19

From this data, the performance of xpu is obviously worse than cpu.
Is there something wrong with my os/kernel/driver/dependencies?
Can you give us some suggestions?

[huiyan2021]

Hi @wenkema , thanks for reporting this issue. For Arc770, please use Data Center driver instead.