官方提供的Docker模拟环境中尝试转换模型成为MLU220的离线推导模型，以下是代码

CNML 7.9.114 06a1920

CNRT 4.7.12 03ea1d9

Python 3.6.8

import torchvision.models as models

import torch

import torch_mlu

import torch_mlu.core.mlu_model as ct

import torch_mlu.core.mlu_quantize as mlu_quantize

net = models.resnet50()

# net = models.quantization.resnet50(quantize=True)

dtype = "int8"

per_channel = False

mean = [0.485, 0.456, 0.406]

std = [0.229, 0.224, 0.225]

net.eval().float()

use_avg = False

data_scale = 1.0

dszie = 224

batch_size = 32

half_input = True

mname = 'resnet50'

qconfig = {'iteration': 32, 'use_avg':use_avg, 'data_scale':data_scale, 'mean': mean, 'std': std, 'per_channel':per_channel}

quantized_weight = mlu_quantize.quantize_dynamic_mlu(net, qconfig, dtype=dtype, gen_quant = True)

quantized_model = mlu_quantize.quantize_dynamic_mlu(net)

quantized_model.load_state_dict(quantized_weight.state_dict())

quantized_model.eval().float()

# prepare input

example_mlu = torch.randn(batch_size, 3, dszie, dszie, dtype=torch.float)

randn_mlu = torch.randn(1, 3, dszie, dszie, dtype=torch.float)

if half_input:

    randn_mlu = randn_mlu.type(torch.HalfTensor)

    example_mlu = example_mlu.type(torch.HalfTensor)

# set offline flag

ct.set_core_number(4)

ct.set_core_version('MLU220')

ct.set_device(-1)

ct.save_as_cambricon(mname)

ct.set_input_format(0)

# run jit fuse

net_traced = torch.jit.trace(quantized_model.to(ct.mlu_device()),

                             randn_mlu.to(ct.mlu_device()),

                             check_trace=False)

# run inference and save cambricon

net_traced(example_mlu.to(ct.mlu_device()))

# unset offline flag

ct.save_as_cambricon("")

运行到net_traced(example_mlu.to(ct.mlu_device()))会出现报错，以下是报错信息

最后程序会出现段错误结束程序，最终会产生MLU220的离线推理文件，但是文件大小存在问题，可能只有几兆