Type error can't pickle _thread.lock object learn.export()

saris.kiat · April 30, 2022, 10:03am

I’m a very newbie here I’m in chapter two which I can’t export my model as in the video
here is an error

why could this happens??

is it because I didn’t download image as the instructor use the Bing API which I can’t so my download image code look like this

is that why I can’t export model??
thank you for your kindly answer

Hadus · May 1, 2022, 4:09pm

I really do not think it is because you downloaded the images differently. If you didn’t edit any other part of the code it should work. Have you tried restarting the whole notebook and running everything from the start?

saris.kiat · May 2, 2022, 2:47am

yeah I did try it but it still can’t export

bobertlo · May 2, 2022, 3:03am

I have been seeing this error also. I also used the bing dataloader and considered that it could be causing the issue, but I have reproduced this in a different notebook (and pip venv) loading the files directly from the directory and training a model.

I am using WSL and pip (with upstream pytorch installation instructions and pip install fastai)

EDIT: I wonder if some dependency snuck in a lock with an update? I could imagine environmental issue in my setup but if it shows up on collab that seems more like a bug?

bobertlo · May 2, 2022, 3:24am

github.com/fastai/fastai

v2.6.2: learn.export() throws an error

opened 07:47AM - 30 Apr 22 UTC

closed 11:33AM - 01 May 22 UTC

janvdp

# Describe the bug Exporting a trained vision model throws an error: ```Py…thon from fastai.vision.all import * path = untar_data(URLs.DOGS) dls = ImageDataLoaders.from_folder(path, seed=12, item_tfms=Resize(460), batch_tfms=aug_transforms(size=297), bs=24) learn = vision_learner(dls, resnet50, metrics=error_rate) learn.model.eval(); learn.export() --------------------------------------------------------------------------- TypeError Traceback (most recent call last) [<ipython-input-22-fa5b61306ef3>](https://localhost:8080/#) in <module>() ----> 1 learn.export() 2 frames [/usr/local/lib/python3.7/dist-packages/torch/serialization.py](https://localhost:8080/#) in _save(obj, zip_file, pickle_module, pickle_protocol) 587 pickler = pickle_module.Pickler(data_buf, protocol=pickle_protocol) 588 pickler.persistent_id = persistent_id --> 589 pickler.dump(obj) 590 data_value = data_buf.getvalue() 591 zip_file.write_record('data.pkl', data_value, len(data_value)) TypeError: can't pickle _thread.lock objects ``` # Setup I have tested this locally and on Google Colab with the following versions installed: PyTorch: 1.11.0 fastcore: 1.4.2 fastai: 2.6.2 # Reproduce ```Python from fastai.vision.all import * import torch print(torch.__version__) import fastcore print(fastcore.__version__) import fastai print(fastai.__version__) path = untar_data(URLs.DOGS) dls = ImageDataLoaders.from_folder(path, seed=12, item_tfms=Resize(460), batch_tfms=aug_transforms(size=297), bs=24) learn = vision_learner(dls, resnet50, metrics=error_rate) # Not needed to reproduce #learn.fit_one_cycle(1) learn.model.eval(); learn.export() ``` Here is a link for an example to run on Colab https://colab.research.google.com/drive/1OV_21L-XzcE7Mmsa8LiZeANQExwMmDX7?usp=sharing

Good news. I dug deeper on github and this issue was fixed in the latest version.

I was actually running in the old venv accidentally and switching to the one I created just now while troubleshooting fixed the problem. If you update your fastai library it should work for you.

saris.kiat · May 2, 2022, 4:02am

thank you so much it’s definitely working