m3hrdadfi / albert-persian

ALBERT: A Lite BERT for Self-supervised Learning of Language Representations for the Persian Language

Hi There 👋 , I'm Mehrdad Farahani

I am currently in my third year of a PhD program at Chalmers University of Technology, which I began in 2022 under the supervision of Richard Johansson (CTH) and Gabriel Skantze (KTH). My current research focuses on the controllability and interpretability of language models. For nearly a year, I have been working in this area.

As part of my goal to advance research in Conversational AI, I have decided to work on small pieces of this large puzzle. Specifically, I aim to understand how current language models used in conversational AI applications can perceive and comprehend the broader aspects of this complex field by focusing more on language models' controllability and interpretability.

You can also follow me on:

• Twitter
• Huggingface
• Linkedin
• Google Scholar
• Researchgate

Research Interests: Natural Language Processing, Representation Learning, Controllability, Interpretability

albert-persian's Issues

request for contribution (I wrote an Example for this repo)

Hi Mehrdad, I wrote an Example for Text Classification with Albert-Persian on Colab
Colab link: https://colab.research.google.com/drive/1ICmlf0zLiomtsFsCkkQny_HJN6X2w_of?usp=sharing
it's like your Taaghche_Sentiment_Analysis on ParsBert
I trained it both on TF 2.0 and PyTorch

it's almost done
it just needs some comments and notations for how to use

please check it, and if is this ok?
accept my PR for adding this on the notebooks folder and examples on the Readme File

note: I didn't send the PR yet

I'm sorry, I forgot to convert it into shareable (Albert Example)

I'm sorry, I forgot to convert it into shareable
new link:https://colab.research.google.com/drive/1ICmlf0zLiomtsFsCkkQny_HJN6X2w_of?usp=sharing
please check this out

Originally posted by @sajjjadayobi in #4 (comment)

some note !!

hi Merhdad, thank you for your great job,
you don't have any TF-model in your model card
as you can't see here https://huggingface.co/m3hrdadfi/albert-fa-base-v2/tree/main

and another point,
as you know that for working with Albert we need to install the sentence piece library
but I think it's would be great if you mention it in the readme file
(in the first use I didn't know that maybe there are some people like me)

how to speed up the prediction?

The fine-tuned this model on your dataset is good in my task, but I found it takes around 200ms to process input (one sentence) using CPU. Is there any suggestion to speed up the prediction? Can the prediction time be less than 20 milliseconds?
Thanks.

use for different entity

Hi
Is it possible to The fine-tuned this model on my dataset that have different entities?
thanks