multiNLI
multiNLI copied to clipboard
nyu-mll
In python/train_snli.py , line 246 The code gives index out of bound error, because `results` is a tuple with 2 elements. ```python results = evaluate_final(classifier.restore, classifier.classify, [test_matched, test_mismatched, test_snli], FIXED_PARAMETERS["batch_size"])...
I downloaded GloVe from [http://nlp.stanford.edu/data/glove.840B.300d.zip](url) and ran: ``` path = '../path_to_glove/file.txt' read_file = open(path, 'r', encoding="utf-8") for i, el in enumerate(read_file): if len(el.split())>301: print(i) print(len(el.split())) ``` it generated: ``` 52343...
For some words in the glove embedding we get the following error: _ValueError: could not broadcast input array from shape (302) into shape (300)_ The following pull request corrects the...
Hello,I am also following esim model these days . I wonder if your multiNLI model can achieve an accuracy of 88.0% in SNLI task according to the paper
@woollysocks : The pull request for loading word vectors effectively.
The repository is quite big and therefore takes a long time to clone. Checking with the `git_find_big.sh` script from https://confluence.atlassian.com/bitbucket/reduce-repository-size-321848262.html shows that this space is taken up by log files....
declared twice. It appears that the version at the bottom is redundant. I haven't checked the other train_*.py files.
