[Help] Error in generation pipeline

[kwcckw]

Hi,

Good day.

I'm facing error below when trying to generate text with ara-gpt2:

And here's the code:

import torch

from transformers import GPT2TokenizerFast, pipeline
#for base and medium
from transformers import GPT2LMHeadModel
#for large and mega
from aragpt2.grover.modeling_gpt2 import GPT2LMHeadModel

from preprocess import ArabertPreprocessor

device = 0 if torch.cuda.is_available() else -1

MODEL_NAME='aragpt2-base'
arabert_prep = ArabertPreprocessor(model_name=MODEL_NAME)

text1  = 'ذهب نوح مظفر ضرغام بصحبة رؤوف بن لؤي ظافر عطعوط وهلال  ';
text2 = ' خازن عفيف للحج';

text_clean = arabert_prep.preprocess(text1)

model = GPT2LMHeadModel.from_pretrained(MODEL_NAME)
tokenizer = GPT2TokenizerFast.from_pretrained(MODEL_NAME)
generation_pipeline = pipeline("text-generation",model=model,tokenizer=tokenizer, device=device)

input_size = len(tokenizer.tokenize(text_clean))

gen_text = generation_pipeline(text_clean,
            pad_token_id=0, # 0 for AraGPT2
            num_beams=5,
            max_length=text_clean+1,
            top_p=0.95,
            repetition_penalty = 3.0,
            no_repeat_ngram_size = 3)[0]['generated_text']

Could you advise on any troubleshooting steps? At this point i'm quite clueless on the error.

Thank you.

[WissamAntoun]

check this notebook with has the code running for all AraGPT2 models https://colab.research.google.com/drive/1T5xKRAtKeuUmmokyThvkK8pulHw02CoK?usp=sharing#scrollTo=hDeZLZxypXGQ

I think the issue in the code above is that you are using the GPT2LMHeadModel class from our repository with the base model, instead of using the one provided by Huggingface

[kwcckw]

Hi Wissam Antoun,

Yes, i did following that colab file and i downloaded the files from Huggingface:

from Huggingface:

The one in local disk:

I'm downloaded the files from this link:
https://huggingface.co/aubmindlab/aragpt2-base/tree/main

If this is not the correct path, could you point me to the correct path?

Thank you.

[WissamAntoun]

Yes this is the correct repository.
To use the model just provide the path of the folder you just downloaded to model parameter in the pipeline class.

[kwcckw]

I tried to redownload the whole repo and run with the sample code, but this time i'm getting another error:

Probably is due to version mismatch issue. Is it possible for you provide the complete list of requirements?

Right now i can see there are only 3 libraries in the requirement file:

    #tensorflow >= 1.11.0, <2.0   # CPU Version of TensorFlow.
  
  
    # tensorflow-gpu  >= 1.11.0, <2.0 # GPU version of TensorFlow.
  
  
    PyArabic
  
  
    farasapy

Thank you.

[kwcckw]

Looks like i resolved the issue by reinstalling pip install transformers==4.2.1, the newer version of transformer will facing error as above.

Thanks again for your help and appreciate it.