znsoftm opened a new issue #7715:
URL: https://github.com/apache/tvm/issues/7715
Windows 10 , LLVM 11
error: on this line: lib.export_library(libpath)
Exception has occurred: TVMError
Traceback (most recent call last):
File "E:\tvm\tvm\src\runtime\dso_library.cc", line 66
TVMError: ---------------------------------------------------------------
An internal invariant was violated during the execution of TVM.
Please read TVM's error reporting guidelines.
More details can be found here:
https://discuss.tvm.ai/t/error-reporting/7793.
---------------------------------------------------------------
Check failed: lib_handle_ != nullptr == false: Failed to load dynamic
shared library ./tvm_output_lib/mobildv2.dll
File "E:\tvm-test\tvm_export_new.py", line 63, in <module>
loaded_lib = tvm.runtime.load_module(libpath)
`import onnx
import os
import platform
import tvm
import tvm.relay as relay
from PIL import Image
import numpy as np
from torchvision import transforms
NEWLIBDIR="./tvm_output_lib/"
sysstr = platform.system()
if(sysstr =="Windows"):
libext=".dll"
else:
libext=".so"
libpath=NEWLIBDIR+"mobildv2"+libext
if not os.path.exists(NEWLIBDIR):
os.makedirs(NEWLIBDIR)
# https://zhuanlan.zhihu.com/p/60981432
onnx_model = onnx.load('mobilenetv2.onnx')
img = Image.open('cat.jpg').resize((244, 244))
# 对图片进行正则化处理，参数来自于官方网站
preprocess = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224,
0.225]),
])
img = preprocess(img)
x = img[np.newaxis, :]
# 这里首先在PC的CPU上进行测试 所以使用LLVM进行导出
targetstring="llvm -mcpu=core-avx2"
target = tvm.target.Target(targetstring)
input_name = 'input' # 这儿的名字可以用netron 查看网络结构，找到input节点的name
shape_dict = {input_name: x.shape}
sym, params = relay.frontend.from_onnx(onnx_model, shape_dict)
print(sym)
# 这里利用TVM构建出优化后模型的信息
with tvm.transform.PassContext(opt_level=2):
lib = relay.build(sym, target=target, target_host=target,params=params)
dtype = 'float32'
from tvm.contrib import graph_runtime
# 下面的函数导出我们需要的动态链接库 地址可以自己定义
print("Output model files")
lib.export_library(libpath)
# -------------至此导出模型阶段已经结束--------
# 接下来我们加载导出的模型去测试导出的模型是否可以正常工作
loaded_lib = tvm.runtime.load_module(libpath)
# 这里执行的平台为CPU
#ctx = tvm.cpu()
ctx = tvm.context(targetstring)
module = graph_runtime.GraphModule(lib["default"](ctx))
module.set_input("input", x) #此处也是输入节点名字
module.run()
out_deploy = module.get_output(0).asnumpy()
'''
def soft_max(z):
t=np.exp(z)
a=np.exp(z)/np.sum(t,axis=1)
return a
'''
print(np.argmax(out_deploy))
`
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
