在使用ipex-llm中的speculative模式报错内存不足，并且无法转到xpu上

在使用ipex-llm中的speculative模式报错

RuntimeError: setStorage: sizes [1, 4, 396, 128], strides [201216, 50304, 128, 1], storage offset 0, and itemsize 4 requiring a storage size of 806400 are out of bounds for storage of size 804864

转到xpu上时报错，RuntimeError: PyTorch is not linked with support for xpu devices

import random
import time
import numpy as np
import torch
from transformers import AutoTokenizer#, AutoModelForCausalLM
from ipex_llm.transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained('/home/models/swy/ouput/secQwen', 
                                             load_in_4bit=True,
                                             device_map='cpu',
                                             speculative=True
                                             ).to('cpu') "无法转换到xpu上"

tokenizer = AutoTokenizer.from_pretrained("/home/models/swy/ouput/secQwen", trust_remote_code=True)
# model = AutoModelForCausalLM.from_pretrained("/home/models/swy/ouput/secQwen", device_map='auto', trust_remote_code=True)

data = [
    {
        "instruction":"你是一个信用卡诈骗案件识别专家，请帮助我分析下面案件是否属于信用卡诈骗案件",
        "input":"2020年4月21日，我刑警大队接到肇东市居民马兴阁报案称：2020年4月20日在肇东市十七道街北万福家园楼下大丰收烧烤店自己家中，接到一个陌生电话问其是否需要贷款，马兴阁说需要贷款，对方让其添加QQ为好友后，让其下载泛华金融APP让其在泛华金融APP上办理贷款业务，在办理贷款业务时被骗走人民币3000元。",
        "output":"该案件不属于诈骗案件但不属于信用卡诈骗案件"
    } ]
    
time__s = []
for da in data:
    question = da['instruction'] + "案件描述如下：" + da['input']
    # question = "你叫什么"
    
    s_time = time.time()
    inputs = tokenizer(question, return_tensors="pt")
    input_ids = inputs["input_ids"]
    print("****************************  开始进行模型预测  ****************************")

    pred = model.generate(input_ids,
                        max_new_tokens=300,
                        # max_new_tokens=1,
                        # return_dict_in_generate=True,
                        # output_scores=True,
                        )
    for input_ids, output_ids in zip(input_ids, pred):
        generated_ids = output_ids[len(input_ids):]


    # generated_ids = [
    #     output_ids[len(input_ids):] for input_ids, output_ids in zip(input_ids, pred)
    # ]
    print("****************************  开始进行decode  ****************************")
    text = tokenizer.decode(generated_ids, skip_special_tokens=True)
    e_time = time.time()
    times = e_time-s_time
    print("执行时间：", times, "秒")
    print(text)
    time__s.append((times,len(text)))
t = 0
l = 0
for t,l in time__s:
    t += t
    l += l
print("模型的平均耗时约为：" + str(l/t) + " 字/s" )