Hi, I am trying to run onnx model on Android and have sharted with the steps like is described there: https://github.com/onnx/tutorials/blob/master/tutorials/PytorchCaffe2MobileSqueezeNet.ipynb

import onnx
import caffe2.python.onnx.backend
from onnx import helper

# Load the ONNX GraphProto object. Graph is a standard Python protobuf object
model = onnx.load("model.onnx")

Unfortinately I receive an error:

---------------------------------------------------------------------------
DecodeError                               Traceback (most recent call last)
<ipython-input-8-0e15f43f99e0> in <module>()
      1 # Load the ONNX GraphProto object. Graph is a standard Python protobuf object
----> 2 model = onnx.load("model.onnx")
      3 

2 frames
/usr/local/lib/python3.6/dist-packages/onnx/__init__.py in _deserialize(s, proto)
     95                          '\ntype is {}'.format(type(proto)))
     96 
---> 97     decoded = cast(Optional[int], proto.ParseFromString(s))
     98     if decoded is not None and decoded != len(s):
     99         raise google.protobuf.message.DecodeError(

DecodeError: Error parsing message

Could you please what could be the issue? I use EN model and Google Colab

Hi, a lot of thanks to your project.

In the README, it says:

Alternatively, you can also load your own model.

Where can I can find models for other languages except English and German? Or could you tell me how to train my own model for other languages step by step? I'm happy to contribute for providing more models.

Thank you, Guangrui Wang

After running into #13, I tried to use the Python bindings instead. It worked, but I noticed that installed version 0.2.2 (saw it didn't match up with the documentation in the README).

After digging into it a little bit, I saw that 0.2.2 was the last release with a platform-agnostic wheel available. All 0.3.x wheels seem to be built specifically for macOS, and are not installable on my Linux/Ubuntu machine.

I'm wondering if there are some easy adjustments that could be made to make publishing wheels for all platforms again possible (or at least Linux/Ubuntu :innocent: )?

Hey, there! I was trying to run the Rust example from the README, but got the following error on a cargo run:

Error: Compat { error: TorchError { c_error: "The following operation failed in the TorchScript interpreter.\nTraceback of TorchScript, serialized code (most recent call last):\n  File \"code/__torch__/torch/nn/quantized/dynamic/modules/rnn.py\", line 195, in __setstate__\n    state: Tuple[Tuple[Tensor, Optional[Tensor]], bool]) -> None:\n    _72, _73, = (state)[0]\n    _74 = ops.quantized.linear_prepack(_72, _73)\n          ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE\n    self.param = _74\n    self.training = (state)[1]\n\nTraceback of TorchScript, original code (most recent call last):\n  File \"/usr/local/lib/python3.6/dist-packages/torch/nn/quantized/dynamic/modules/rnn.py\", line 29, in __setstate__\n    @torch.jit.export\n    def __setstate__(self, state):\n        self.param = torch.ops.quantized.linear_prepack(*state[0])\n                     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE\n        self.training = state[1]\nRuntimeError: Didn\'t find engine for operation quantized::linear_prepack NoQEngine\n" } }

Let me know it there is any more info you need for debugging!

After installing the new version, I get the following exception when running NNSplit.load("en").

PanicException: called `Result::unwrap()` on an `Err` value: PyErr { type: Py(0x5632cc1b1140, PhantomData) }

This occurs both with and without onnxruntime-gpu installed.

onnxruntime has deprecated creating a session without setting a provider at creation time (the latest version fails). This PR fixes the session creation.

Hi, I was trying the simple example in Python from the documentation and I'm getting an ImportError:

from nnsplit import NNSplit
splitter = NNSplit.load("en")

# returns `Split` objects
splits = splitter.split(["This is a test This is another test."])[0]

# a `Split` can be iterated over to yield smaller splits or stringified with `str(...)`.
for sentence in splits:
   print(sentence)

When executing this example I'm getting the following error:

Traceback (most recent call last):
  File "nnsplit.py", line 1, in <module>
    from nnsplit import NNSplit
  File "G:\OneDrive\projects\s\nnsplit.py", line 1, in <module>
    from nnsplit import NNSplit
ImportError: cannot import name 'NNSplit' from partially initialized module 'nnsplit' (most likely due to a circular import) (G:\OneDrive\projects\s\nnsplit.py)

I have installed the packages in a new conda environment, executing pip list installed I have:

pip list installed
Package         Version
--------------- -------------------
certifi         2020.12.5
nnsplit         0.5.7.post0
numpy           1.20.3
onnxruntime     1.7.0
onnxruntime-gpu 1.7.0
pip             21.1.1
protobuf        3.17.1
setuptools      52.0.0.post20210125
six             1.16.0
tqdm            4.61.0
wheel           0.36.2
wincertstore    0.2

I'm trying to import nnsplit in a JavaScript project, and webpack is failing with:

./node_modules/nnsplit/nnsplit.bundle/nnsplit_javascript_bg.wasm
Module not found: Can't resolve './nnsplit_javascript_bg.js' in '/tmp/experiment/node_modules/nnsplit/nnsplit.bundle'

Looking in node_modules/nnsplit/nnsplit.bundle, indeed the file nnsplit_javascript_bg.js is referenced by package.json, but missing from the filesystem.

(Not sure though whether that's the real culprit, as the nodejs example seems to work as intended.)

When trying to install into python3.9 it will not install a version later than 0.2.2. I am not certain but I believe that this is because wheels are only build for versions 3.6, 3.7 and 3.8. Would it be possible to add wheels for the 3.9 version?

Hello and first of all: thank you for a great library!

I've tried to train my own model using an unusual input data format following the train Python notebook you've provided. However, after the training, when trying to load the custom model via NNSplit.load("en/model.onnx") call in python bindings, I get this:

nnsplit.ResourceError: model not found: "en/model.onnx"

I may be wrong, but it seems the current logic of model_loader.rs does not allow custom local paths, only the ones that are listed in the models.csv:

https://github.com/bminixhofer/nnsplit/blob/a5a15815382029bf5c3438fd4753f644847d4dbf/nnsplit/src/model_loader.rs#L59

Effectively limiting the available models to the pretrained ones.

This security vulnerability:

https://rustsec.org/advisories/RUSTSEC-2021-0073.html

is fixed in prost==0.8.0, which is included in a recent new release of tract-onnx: https://github.com/sonos/tract/releases/tag/0.15.2

Would it be possible to do a new release with the tract-onnx dependency bumped?

I was trying to segment sentences for my transcribing program, but I ran into this error when I first tried using it this.

Full Error

Traceback (most recent call last):
  File "/home/runner/Voice-Synthasizer/venv/lib/python3.8/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 280, in __init__
    self._create_inference_session(providers, provider_options)
  File "/home/runner/Voice-Synthasizer/venv/lib/python3.8/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 307, in _create_inference_session
    sess = C.InferenceSession(session_options, self._model_path, True, self._read_config_from_model)
RuntimeError: /onnxruntime_src/onnxruntime/core/platform/posix/env.cc:142 onnxruntime::{anonymous}::PosixThread::PosixThread(const char*, int, unsigned int (*)(int, Eigen::ThreadPoolInterface*), Eigen::ThreadPoolInterface*, const onnxruntime::ThreadOptions&) pthread_setaffinity_np failed


During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "speech.py", line 436, in <module>
    transcript, source_align_data = transcript_audio(input_path, True, transcript_path, granularity=granularity)
  File "speech.py", line 271, in transcript_audio
    sentence_segmenter = NNSplit.load("en")
  File "backend.py", line 6, in create_session
  File "/home/runner/Voice-Synthasizer/venv/lib/python3.8/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 283, in __init__
    print("EP Error using {}".format(self._providers))
AttributeError: 'InferenceSession' object has no attribute '_providers'
Segmentation fault (core dumped)

Hello, first of all, nnsplit is really cool, it's really great stuff. :) I'd really like to run nnsplit on my local computer, but an error occurs when I try to pip install nnsplit:

ERROR: Could not find a version that satisfies the requirement nnsplit (from versions: 0.0.1, 0.1.0, 0.1.1, 0.1.2, 0.1.3, 0.1.4, 0.2.0, 0.2.1, 0.2.2) ERROR: No matching distribution found for nnsplit

can I get some helps?

Hi, This is more of a minor feature request. I'm trying to use NNSplit in a container, which has a read-only file system except for the /tmp dir. It would be groovy if one could provide a local path to load the model from/download to. Perhaps this is in the python interface already but i couldn't see it.

I know you can specify a path when calling NNSplit() but this gets more complcated as I'm including it in a modele that then gets included in another project.

Anyway, nice work and thanks!

i run it on centos7 and python3.8.3，and i just want to run it on cpu not gpu. I met follows error：

>>> import nnsplit
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: /lib64/libm.so.6: version `GLIBC_2.27' not found (required by /home/zyb/miniconda3/lib/python3.8/site-packages/nnsplit.cpython-38-x86_64-linux-gnu.so)

Hi,

I have tried Simplified Chinese model on demo page and it seems that sentence boundary and tokens detection are not correct.

I have 2 ideas why that could happen:

1. Period in Chinese is 。
2. There are no white spaces between words. Possibly it is better to use something like https://github.com/voidism/pywordseg to split on words as a preprocessing step

It looks like issue 2. causes that tokens are also not detected correctly. I have compared with https://github.com/voidism/pywordseg results and they do not match. But I am not sure here, because I have compared Spacy, pywordseg and Stanford Word Segmenter and all of them provide different results