hooshvare/parsgpt

Persian GPT2

parsgpt's Issues

how to remove limitation in generated text length

hey im using your model and when i set num_sents to a number bigger than two , is see this
"Setting pad_token_id to eos_token_id:5 for open-end generation."
how and where can I adjust this limitation?
tnx

Fine tuning Persian GPT-2 for Question Answering Task

Is it possible to fine-tune the general Persian GPT-2 model on the question answering task with a custom dataset? I couldn't find any sources describing how to fine-tune a GPT model on a custom dataset. (except using the gpt-2-simple library which doesn't support Farsi or other huggingface models)

ParsGpt comparison with English versions

Hi,
Thank you for providing this useful resource. I would like to know more about the characteristics of this transformer compared to English ones. For example, how many parameters it has and on home many texts it was trained or even the genre and types of text. If possible compare these statistics with a model in English.

Thank you!

Byte-Pair Encoding (BPE) in Persian

Hello,
Why do we have strange characters when we run BPE on the Persian language?

GPT ecco visualization not working

Hi,

Firstly thank you very much for creating such amazing models and explanations of each.

Secondly, I had to report a bug. In the Persian GPT2 - Visualization.ipynb notebook, after running the second cell (lm = ecco.from_pretrained('HooshvareLab/gpt2-fa')) we get the following error:

The model 'HooshvareLab/gpt2-fa' is not defined in Ecco's 'model-config.yaml' file and so is not explicitly supported yet.

I'd really appreciate it if you guys could take a look at this issue. The Ecco library can be a lot of help in understanding this model.

Thanks!