gt4sd / zero-shot-bert-adapters Goto Github PK

Getting this error after a clean install and running sample Jupyter Notebook. On macOS, using vscode in anaconda environment. Happens after running the cell beginning with "model_name = "

The model link you provided cannot be accessed at all? Can you directly feed the model into the code? Or explain what kind of pre trained model is useless? Thank you.

https://github.com/GT4SD/zberta/blob/1968b40ca9ab1ee5f4e65b0d295f80ab18f4e0bb/src/zberta/intent_discovery/zberta.py#L27

Since the method is used also in the example, I would consider removing the prepended _.

After installing all I run the example

model_name = "bert-base-uncased"
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
berta = model.instantiate_model(2, 2, device, model_name,
                                {"entailment": 0, "contradiction": 1}, path="model.pt", dict=True)

and I get error

AttributeError: 'BertModel' object has no attribute 'add_adapter'

Python version: Python 3.10.6
OS Version: `MacOS Monterey 12.6'

I tried to install the package on a clean env using the following commands:

poetry add  git+https://github.com/GT4SD/zberta.git

and

poetry add git+ssh://[email protected]/github.com/GT4SD/zberta.git

and I getting the following error:

CalledProcessError

  Command '['git', 'clone', '--recurse-submodules', '--', 'ssh://[email protected]/github.com/GT4SD/zberta.git', '/var/folders/5t/lgg8r5ws1xbdm8pp6kh37ytm0000gn/T/pypoetry-git-zbertaeirek49a']' returned non-zero exit status 128.

  at /usr/local/Cellar/poetry/1.1.15/libexec/lib/python3.10/site-packages/poetry/utils/_compat.py:217 in run
      213│                 process.wait()
      214│                 raise
      215│             retcode = process.poll()
      216│             if check and retcode:
    → 217│                 raise CalledProcessError(
      218│                     retcode, process.args, output=stdout, stderr=stderr
      219│                 )
      220│         finally:
      221│             # None because our context manager __exit__ does not use them.

python version: 3.9 & 3.10
os version: MacOS Monterey 12.6
poetry version: 1.1.15

Hi, Thank you for your making this pipeline open source. I really appreciate that.

I am having an issue in BERTA instance while training. Please find the error and the code.

Any solution/advice would be really appreciated. Thanks

Hi @comidan, Thanks again for making this pipeline open-source.
I have two concerns.

1. Whether the model is using both datasets (SNLI and Banking77-OOS) or it’s only using SNLI for the training? if yes, then can you please point out where and how is it used in the code?

2. It is related to the 1st one as well. Actually, I need to train the BERT-A model with any other dataset other than Banking77-OOS to generate unknown intents. Can you please guide me on how to do that?

---> 15 z_intents = unknown_intents_set("en_core_web_trf", z_dataset['test']['text'])
16 berta = model.instantiate_model(z_banking.labels(), z_banking.output_dim(), device, model_name,
17 z_banking.nli_labels(), path="model.pt", dict=True)

/usr/local/lib/python3.7/dist-packages/zberta/intent_discovery/unknown_intents.py in unknown_intents_set(dataset, spacy_model)
15 def unknown_intents_set(dataset, spacy_model=None):
16 global nlp_engine
---> 17 if spacy_model is not None and spacy_model not in default_spacy:
18 os.system("python -m spacy download " + spacy_model + " --no-deps")
19 nlp_engine = spacy.load(spacy_model)

TypeError: 'in ' requires string as left operand, not list

When the zberta.find_new_intents function is called on the utterances the following is the output