plantl-gob-es/lm-spanish

Official source for spanish Language Models and resources made @ BSC-TEMU within the "Plan de las Tecnologías del Lenguaje" (Plan-TL).

lm-spanish's Issues

GPT-2 state and GPT-j-6B

I would like to ask about the state of the GPT-2 model. Will it arrive soon at huggingface?

I would also like to ask if you have the intention of train GPT-j-6B. Training this model for some people would be impossible due to its hardware requirements, but you have Mare Nostrum, the dataset and the previous version GPT-2.

Tokenizer rare subwords

Hi, First at all many thanks for this amazing work in Spanish 😍.

from transformers import AutoModelForMaskedLM
from transformers import AutoTokenizer

model_checkpoint = "PlanTL-GOB-ES/roberta-large-bne"
model = AutoModelForMaskedLM.from_pretrained(model_checkpoint)
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)

rare_tokens_vocab = [word for word in tokenizer.vocab if 'Ġ' in word]
print(len(rare_tokens_vocab))
# 37695 50262

Almost 75% or tokens contains "Ġ" char. It is really strange! Probably due to dirty text in the corpus ?? or why is the reason of so many tokens with "Ġ"??

Potential issues with HF GPT2 Models

Hello,

I am using the GPT2 models available in HF, and running into a few issues. Firstly, there seems to be an issue with the tokenizer. Trying to calculate perplexity using the evaluate module, as follows:

from evaluate import load
perplexity = load("perplexity", module_type="metric")
results = perplexity.compute(predictions=["Hola, como estas?"], model_id="PlanTL-GOB-ES/gpt2-base-bne", device="cpu")

Gives the following error:

 ...
  File "/ikerlariak/aormazabal024/PhD/Poetry-Generation/demo/poetry-env-traganarru/lib/python3.8/site-packages/torch/nn/functional.py", line 2199, in embedding
    return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
IndexError: index out of range in self`

This seems to be related to the special tokens for <pad>, <s>, </s> and <unk> not being properly set (but are used by the evaluate module), as the only special token added in the tokenizer is <|endoftext|>. One can manually fix it for the local snapshot:

tokenizer.pad_token = '<pad>'
tokenizer.bos_token = '</s>'
tokenizer.eos_token = '</s>'
tokenizer.unk_token = '<unk>'
tokenizer.save_pretrained('[snapshot-path]')

However, even after fixing this, I am getting quite high perplexities compared to the 10-13 reported in the paper for all sentences I am trying (assuming per-word-perplexity is reported). Is it possible there was an issue when converting from fairseq to HF, and are the original fairseq models available somewhere to compare? Or maybe I am making a mistake when calculating the ppl, was there any tokenization done to the text apart from BPE (i.e. replacing newlines with , which is pretty standard in fairseq)?

Dataset Availability

Hi!

Thanks for the work, I love it. I was wandering if the final clean corpus is available somewhere.

Agila dataset availability

Hi! Thanks for all the work!

I would like to ask if the dataset used for the training of the agila-7b model is available for testing/downloading.