How to convert PyTorch model to Tensorflow

In this video, we will convert the Pytorch model to Tensorflow using (Open Neural Network Exchange) ONNX. ONNX is a standard format supported by a community of partners such as Microsoft, Amazon, and IBM. PyTorch is mainly maintained by Facebook and Tensorflow is built in collaboration with Google.

Repository

Android application:

One year of unlimited online learning. Get 30% OFF Coursera Plus.

Expanda seus horizontes de carreira. Tenha 68% de desconto por um ano de Coursera Plus

Learn from experts at Google and get in-demand AI skills you can apply to your work right away with Google AI Essentials, with zero experience required.

Boost your software development career with Gen AI. Build in-demand, hands-on, Generative AI skills for elevating your software engineering game in 1 month or less.

Join the channel to get access to more perks:

Or visit my blog at:

Outro music: Sanaas Scylla

#pytorch #tensorflow #machinelearning

[kidus]

Hello Daniel, I am having this error
AttributeError: type object 'TFLiteConverterV2' has no attribute 'from_frozen_graph'

[H-uu]

Would it be applicable for a model which is traned in Resnet50 (which uses Pytorch) to tensorflowlite

[hammadmuhammad]

Hello Daniel, Thanks for uploading this video on a guide to convert pytorch model to tflite, But actually i am stuck in a situation where i want to convert a Voice recognition model from Hugging face which is in availible in pytorch format and i want to convert it into tensorflow lite for my Mobile app, Can you please get the same concept and make a video of converting a voice recognition model from pytorch to tensorflow lite. I will be Very very thankfull for that pleasss

[chandreshyadav]

Hello Daniel . I am using Checkpoints in my model but the given checkpoint is in .pth extension whereas model has .bin checkpoint file.. how can I use .pth as checkpoint in my model??

[sonalibhargav]

Hi.. Daniel.. I want to convert one pytorch model in tensorflow but i don't have the architecture of model.. I have just the model. I use the jit for load the model but I am not able to convert the model. Can you please help?

[supremee]

Hi Daniel, is there any way to convert a model from pytorch model to tensor flow check point (.ckpt, .meta, .index)

[sondoongee]

Hello Daniel. Thanks for great video.

Is there any way to not freeze the model when coverted to tensorflow? I need ckpt file but the module only produces pb file.

[anandkrishnan]

I keep getting this error when I try and convert my py torch file:

Traceback (most recent call last):
File "pytorch_to_tensorflow_lite.py", line 44, in <module>
trained_dict = torch.load(sys.argv[1], map_location={'cuda:0': 'cpu'})
File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/torch/serialization.py", line 584, in load
return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args)
File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/torch/serialization.py", line 842, in _load
result = unpickler.load()
ModuleNotFoundError: No module named 'models.yolo'

[shubhamthapa]

Hey @
Daniel Persson can you confirm that this method is still working in 2021 ?

[amishajain]

hi Daniel i got the error
FileNotFoundError Traceback (most recent call last)
in <module>
1 # Load the trained model from file
----> 2 trained_dict = torch.load(sys.argv[1], map_location={'cuda:0': 'cpu'})
3
4 trained_model = MLP(784, [256, 256], 10)
5

in load(f, map_location, pickle_module, **pickle_load_args)
577 pickle_load_args['encoding'] = 'utf-8'
578
--> 579 with _open_file_like(f, 'rb') as opened_file:
580 if _is_zipfile(opened_file):
581 # The zipfile reader is going to advance the current file position.

in _open_file_like(name_or_buffer, mode)
228 def _open_file_like(name_or_buffer, mode):
229 if _is_path(name_or_buffer):
--> 230 return _open_file(name_or_buffer, mode)
231 else:
232 if 'w' in mode:

in __init__(self, name, mode)
209 class _open_file(_opener):
210 def __init__(self, name, mode):
--> 211 super(_open_file, self).__init__(open(name, mode))
212
213 def __exit__(self, *args):

FileNotFoundError: [Errno 2] No such file or directory: '-f'

[prabhatale]

hey, daniel...is there a way to convert pytorch model to .h5 format