marcopoli / alberto-it Goto Github PK

Hi! Thank you for the brilliant work! I am using Italian tweets for my research and want to use AlBERTo-it for sentiment analysis. We have hand-coded training set and hope to fine-tune the model. I am wondering if there's any type of performance score (such as probability, confidence, etc) so that I could conduct a interactive process in fine-tuning the model. I found such description in your paper, however didn't find related codes mentioned in github.

More guidance would be Much appreciated!

Hi all,

Thanks for the cool contribution. I looked at the paper and found information on preprocessing but no mention of sentence splitting. You didn't do it then or did I miss something?

Thanks a lot in advance.

Hi. I need to use Tensorflow 2 but seems that alBERTo is compatible only with Tensorflow 1.x. Someone could help me?

Hi dear I can't load the example with your indications. Where I 'm failing?

from tokenizer import *
from transformers import AutoTokenizer, AutoModel

a = AlBERTo_Preprocessing(do_lower_case=True)
s: str = "#IlGOverno presenta le linee guida sulla scuola #labuonascuola - http://t.co/SYS1T9QmQN"
b = a.preprocess(s)

tok = AutoTokenizer.from_pretrained("m-polignano-uniba/bert_uncased_L-12_H-768_A-12_italian_alb3rt0")
tokens = tok.tokenize(b)
print(tokens)

model = AutoModel.from_pretrained("m-polignano-uniba/bert_uncased_L-12_H-768_A-12_italian_alb3rt0")

When I write a = AlBERTo_Preprocessing(do_lower_case=True), I have an "unresolved reference AlBERTo_Preprocessing". What I could do? Thank you dear

I would like to use Alberto's pretrained model with Huggingface's Transformer package.
Can you provide the model in a format that's compatible/loadable with that package?

Hi, I have a question about the training set of AlBERTo.
I've read that the pre-trained lower cased model is based on 200M of tweets (191GB of raw data) but what about the training set only? Could you please specify how large the training set is?
Thanks a lot.