search

AIPHES / emnlp19-moverscore

MoverScore: Text Generation Evaluating with Contextualized Embeddings and Earth Mover Distance
MIT License
197 stars 34 forks source link

a little question #5

Closed eyuansu62 closed 4 years ago

eyuansu62 commented 4 years ago

You mentioned you fine-tuned the Bert model on some NLI tasks, and then used its model to embedded the sentences. But the code you released did not contain the related code and fine-tuned Bert model, right?

andyweizhao commented 4 years ago

I did release the BERT model fine-tuned on NLI. Please check out line 24 and line 65 in moverscore.py, but I did not release the code about how to fine-tune BERT because you can find it in https://github.com/huggingface/transformers.

eyuansu62 commented 4 years ago

Oh, i see, thank you!! And in "model = BertForSequenceClassification.from_pretrained(output_dir, 3)", you set the labels=3, why the label is set to 3?

andyweizhao commented 4 years ago

MNLI corpus has three labels: neutral, contradiction and entailment.

eyuansu62 commented 4 years ago

Oh, i understand! Really really thanks!!!

eyuansu62 commented 4 years ago

hi，can I ask you another question ？ Why the the score obtained by end is always between 0 and 1, except for those informal cases？ It is very kind for your reply

---Original--- From: "Wei Zhao"<notifications@github.com> Date: Mon, Dec 2, 2019 17:28 PM To: "AIPHES/emnlp19-moverscore"<emnlp19-moverscore@noreply.github.com>; Cc: "Author"<author@noreply.github.com>;"eyuansu62"<772468951@qq.com>; Subject: Re: [AIPHES/emnlp19-moverscore] a little question (#5)

MNLI corpus has three labels: neutral, contradiction and entailment.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.