Vision: Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 2. Your PyTorch installation may be too old. (init at ../caffe2/serialize/inline_container.cc:132)

Created on 5 Mar 2020  路  18Comments  路  Source: pytorch/vision

I have created .pt file of keypoint_rcnn using torch=1.5 and vision=0.6.0 now trying to integrate it with my android app. The error I am getting is as below.

E/AndroidRuntime: FATAL EXCEPTION: main
Process: org.pytorch.helloworld, PID: 15691
java.lang.RuntimeException: Unable to start activity ComponentInfo{org.pytorch.helloworld/org.pytorch.helloworld.MainActivity}: com.facebook.jni.CppException: version_ <= kMaxSupportedFileFormatVersion INTERNAL ASSERT FAILED at ../caffe2/serialize/inline_container.cc:132, please report a bug to PyTorch. Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 2. Your PyTorch installation may be too old. (init at ../caffe2/serialize/inline_container.cc:132)
(no backtrace available)
at android.app.ActivityThread.performLaunchActivity(ActivityThread.java:3122)
at android.app.ActivityThread.handleLaunchActivity(ActivityThread.java:3261)
at android.app.servertransaction.LaunchActivityItem.execute(LaunchActivityItem.java:78)
at android.app.servertransaction.TransactionExecutor.executeCallbacks(TransactionExecutor.java:108)
at android.app.servertransaction.TransactionExecutor.execute(TransactionExecutor.java:68)
at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1977)
at android.os.Handler.dispatchMessage(Handler.java:106)
at android.os.Looper.loop(Looper.java:193)
at android.app.ActivityThread.main(ActivityThread.java:6923)
at java.lang.reflect.Method.invoke(Native Method)
at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:493)
at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:870)
Caused by: com.facebook.jni.CppException: version_ <= kMaxSupportedFileFormatVersion INTERNAL ASSERT FAILED at ../caffe2/serialize/inline_container.cc:132, please report a bug to PyTorch. Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 2. Your PyTorch installation may be too old. (init at ../caffe2/serialize/inline_container.cc:132)
(no backtrace available)
at org.pytorch.NativePeer.initHybrid(Native Method)
at org.pytorch.NativePeer.(NativePeer.java:18)
at org.pytorch.Module.load(Module.java:23)
at org.pytorch.helloworld.MainActivity.onCreate(MainActivity.java:41)
at android.app.Activity.performCreate(Activity.java:7148)
at android.app.Activity.performCreate(Activity.java:7139)
at android.app.Instrumentation.callActivityOnCreate(Instrumentation.java:1293)
at android.app.ActivityThread.performLaunchActivity(ActivityThread.java:3102)
... 11 more

models needs discussion mobile object detection
Source
picture hurricane013

Most helpful comment

If anyone encounters this, this is what happened to me:

  1. Updated torchvision to a version not compatible with torch
  2. Run code that loaded a torchvision dataset, then threw an error
  3. Downgraded torchvision to proper version
  4. Re-ran code that loaded that dataset, which had been downloaded in a newer format, which couldn't be recognized by older torchvision

To fix it, I removed the dataset that torchvision had already downloaded.

picture jxmorris12 on 30 Jul 2020
👍5 ❤2

All 18 comments

Hi,

I suppose you are trying to use torchscript to convert the model to mobile, is that correct?

I suspect that this might not work out-of-the-box currently, because PyTorch mobile doesn't yet support using operators registered outside of PyTorch (like roi_align in torchvision).

@dreiss is that still the case?

fmassa picture fmassa on 10 Mar 2020

Yes, but I think this issue is unrelated. It's just due to a change in the serialization format (which I think was reverted).

dreiss picture dreiss on 10 Mar 2020
👍6

Ok, following @dreiss comment, if you update PyTorch and torchvision this issue should go away, but you'll still have issues exporting keypoint-rcnn to mobile.

fmassa picture fmassa on 10 Mar 2020

Yes @fmassa you are correct. I used pytorch 1.4 snapshot as a dependency, above error disappeared. But still I am getting build failed that even without showing any error. Now I am confused with what should be exact issue.

hurricane013 picture hurricane013 on 11 Mar 2020
❤1 👍1

@hurricane013 as I mentioned in the above post, we currently do not support exporting keypoint-rcnn to mobile yet unfortunately, but support for it will be added in future releases.

fmassa picture fmassa on 11 Mar 2020
👍1

Hi, I wrote a simple MNIST classifier with CNN and got the same error on my android project.

E/AndroidRuntime: FATAL EXCEPTION: main Process: com.example.predictor, PID: 7253 java.lang.RuntimeException: Unable to start activity ComponentInfo{com.example.predictor/com.example.predictor.MainActivity}: com.facebook.jni.CppException: version_ <= kMaxSupportedFileFormatVersion INTERNAL ASSERT FAILED at ../caffe2/serialize/inline_container.cc:132, please report a bug to PyTorch. Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 2. Your PyTorch installation may be too old. (init at ../caffe2/serialize/inline_container.cc:132) (no backtrace available) at android.app.ActivityThread.performLaunchActivity(ActivityThread.java:3270) at android.app.ActivityThread.handleLaunchActivity(ActivityThread.java:3409) at android.app.servertransaction.LaunchActivityItem.execute(LaunchActivityItem.java:83) at android.app.servertransaction.TransactionExecutor.executeCallbacks(TransactionExecutor.java:135) at android.app.servertransaction.TransactionExecutor.execute(TransactionExecutor.java:95) at android.app.ActivityThread$H.handleMessage(ActivityThread.java:2016) at android.os.Handler.dispatchMessage(Handler.java:107) at android.os.Looper.loop(Looper.java:214) at android.app.ActivityThread.main(ActivityThread.java:7356) at java.lang.reflect.Method.invoke(Native Method) at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:492) at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:930) Caused by: com.facebook.jni.CppException: version_ <= kMaxSupportedFileFormatVersion INTERNAL ASSERT FAILED at ../caffe2/serialize/inline_container.cc:132, please report a bug to PyTorch. Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 2. Your PyTorch installation may be too old. (init at ../caffe2/serialize/inline_container.cc:132) (no backtrace available) at org.pytorch.NativePeer.initHybrid(Native Method) at org.pytorch.NativePeer.<init>(NativePeer.java:18) at org.pytorch.Module.load(Module.java:23) at com.example.predictor.MainActivity.onCreate(MainActivity.kt:47) at android.app.Activity.performCreate(Activity.java:7802) at android.app.Activity.performCreate(Activity.java:7791) at android.app.Instrumentation.callActivityOnCreate(Instrumentation.java:1299) at android.app.ActivityThread.performLaunchActivity(ActivityThread.java:3245) ... 11 more Disconnected from the target VM, address: 'localhost:8602', transport: 'socket'

cansozbir picture cansozbir on 5 Apr 2020
👍2

@hurricane013 simple CNN models should work with mobile, I would expect this to be a version issue and if you update PyTorch / torchvision and re-save your weights, it should work

fmassa picture fmassa on 6 Apr 2020

I see.
These are my package versions:

import torch
torch.__version__
'1.5.0.dev20200306'
import torchvision
torchvision.__version__
'0.6.0a0+f534fb5'

and this is what I added to dependencies.

implementation 'org.pytorch:pytorch_android:1.4.0'
implementation 'org.pytorch:pytorch_android_torchvision:1.4.0'

Should I install pytorch's 1.4 version or is there any implementation for 1.5?

cansozbir picture cansozbir on 6 Apr 2020

Yes, I think you might want to install PyTorch 1.4 and torchvision 0.5.0.

But I would recommend posting this question on the pytorch discuss forums, as this is outside of the scope of torchvision and I don't have much context to be able to help with android-specific issues

fmassa picture fmassa on 6 Apr 2020

@hurricane013 simple CNN models should work with mobile, I would expect this to be a version issue and if you update PyTorch / torchvision and re-save your weights, it should work

@fmassa As I mentioned, I used 1.4 snapshot which is latest as a dependency in my android app, but i am still getting build failed without any error.

hurricane013 picture hurricane013 on 25 Apr 2020

If anyone encounters this, this is what happened to me:

  1. Updated torchvision to a version not compatible with torch
  2. Run code that loaded a torchvision dataset, then threw an error
  3. Downgraded torchvision to proper version
  4. Re-ran code that loaded that dataset, which had been downloaded in a newer format, which couldn't be recognized by older torchvision

To fix it, I removed the dataset that torchvision had already downloaded.

jxmorris12 picture jxmorris12 on 30 Jul 2020
👍5 ❤2

@jxmorris12 maybe doing new colab env is much easier...

It just happend to me....so ANNOYING!!
What version of torch, torchvision is needed....???!
EDIT : PyTorch 1.4 and torchvision 0.5.0 seems to work but I'm not perfectly sure yet, and maybe there is more good version to safe. I don know latest version of org.pytorch:pytorch_android yet also.

SungmanHong picture SungmanHong on 11 Sep 2020
👍1

I'm having the same issues on torch=1.4.0 and torchvision=0.5.0. Here's my error message:

Traceback (most recent call last):
File "train.py", line 15, in
model.setup(opt)
File "/home/cchoi/n2d/models/base_model.py", line 29, in setup
self.load_networks(opt.epoch)
File "/home/cchoi/n2d/models/base_model.py", line 129, in load_networks
state_dict = torch.load(load_path, map_location=str(self.device))
File "/home/cchoi/anaconda3/lib/python3.8/site-packages/torch/serialization.py", line 527, in load
with _open_zipfile_reader(f) as opened_zipfile:
File "/home/cchoi/anaconda3/lib/python3.8/site-packages/torch/serialization.py", line 224, in __init__
super(_open_zipfile_reader, self).__init__(torch._C.PyTorchFileReader(name_or_buffer))
RuntimeError: version_ <= kMaxSupportedFileFormatVersion INTERNAL ASSERT FAILED at /opt/conda/conda-bld/pytorch_1579022027550/work/caffe2/serialize/inline_container.cc:132, please report a bug to PyTorch. Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 2. Your PyTorch installation may be too old. (init at /opt/conda/conda-bld/pytorch_1579022027550/work/caffe2/serialize/inline_container.cc:132)
frame #0: c10::Error::Error(c10::SourceLocation, std::string const&) + 0x47 (0x7f0db93f5627 in /home/cchoi/anaconda3/lib/python3.8/site-packages/torch/lib/libc10.so)
frame #1: caffe2::serialize::PyTorchStreamReader::init() + 0x1f5b (0x7f0dbde669ab in /home/cchoi/anaconda3/lib/python3.8/site-packages/torch/lib/libtorch.so)
frame #2: caffe2::serialize::PyTorchStreamReader::PyTorchStreamReader(std::string const&) + 0x64 (0x7f0dbde67bc4 in /home/cchoi/anaconda3/lib/python3.8/site-packages/torch/lib/libtorch.so)
frame #3: + 0x6ce4b6 (0x7f0deba224b6 in /home/cchoi/anaconda3/lib/python3.8/site-packages/torch/lib/libtorch_python.so)
frame #4: + 0x28b8a7 (0x7f0deb5df8a7 in /home/cchoi/anaconda3/lib/python3.8/site-packages/torch/lib/libtorch_python.so)

frame #35: __libc_start_main + 0xf0 (0x7f0dfb530840 in /lib/x86_64-linux-gnu/libc.so.6)

chinglamchoi picture chinglamchoi on 2 Oct 2020
👍1

I ran into the same problem with pytorch 1.4

eric4note picture eric4note on 9 Oct 2020

The model you are trying to load is trained on PyTorch V1.6 or above. Upgrade your PyTorch version to 1.6.

neeraj-j picture neeraj-j on 11 Oct 2020
👍5

Does anyone know how I can reshape Tensor object on android?

Emmanuel12345678 picture Emmanuel12345678 on 30 Oct 2020

The model trained with PyTorch 1.7 and the backbone is resnet50 , but PyTorch version 1.4 loading the model occurred the same error

ForrestPi picture ForrestPi on 6 Nov 2020
👍5

There could be other serious bugs inside.

yechengxi picture yechengxi on 10 Nov 2020
Was this page helpful?
0 / 5 - 0 ratings

Related issues

alpha-gradient picture alpha-gradient  路  3Comments
varagrawal picture varagrawal  路  3Comments
martinarjovsky picture martinarjovsky  路  4Comments
Wadaboa picture Wadaboa  路  3Comments
Linardos picture Linardos  路  4Comments