Edit model card

Google's mT5-XL - Finetuned for Hebrew Question-Answering

Google's mT5 multilingual Seq2Seq model, finetuned on HeQ for the Hebrew Question-Answering task.

This is the model that was reported in the DictaBERT release here.

Sample usage:

import torch
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained('dicta-il/mt5-xl-heq')
model = AutoModelForSeq2SeqLM.from_pretrained('dicta-il/mt5-xl-heq')

model.eval()

question='ื›ื™ืฆื“ ื”ื•ื’ื‘ืœ ื”ืžื™ื“ืข ืฉื ื™ืชืŸ ืœื”ืฉื™ื’ ื‘ืืžืฆืขื•ืช ื”ืขื•ื’ื™ื•ืช?'
context='ื‘ื ื™ื™ืช ืคืจื•ืคื™ืœื™ื ืฉืœ ืžืฉืชืžืฉื™ื ื ื—ืฉื‘ืช ืขืœ ื™ื“ื™ ืจื‘ื™ื ื›ืื™ื•ื ืคื•ื˜ื ืฆื™ืืœื™ ืขืœ ื”ืคืจื˜ื™ื•ืช. ืžืกื™ื‘ื” ื–ื• ื”ื’ื‘ื™ืœื• ื—ืœืง ืžื”ืžื“ื™ื ื•ืช ื‘ืืžืฆืขื•ืช ื—ืงื™ืงื” ืืช ื”ืžื™ื“ืข ืฉื ื™ืชืŸ ืœื”ืฉื™ื’ ื‘ืืžืฆืขื•ืช ืขื•ื’ื™ื•ืช ื•ืืช ืื•ืคืŸ ื”ืฉื™ืžื•ืฉ ื‘ืขื•ื’ื™ื•ืช. ืืจืฆื•ืช ื”ื‘ืจื™ืช, ืœืžืฉืœ, ืงื‘ืขื” ื—ื•ืงื™ื ื ื•ืงืฉื™ื ื‘ื›ืœ ื”ื ื•ื’ืข ืœื™ืฆื™ืจืช ืขื•ื’ื™ื•ืช ื—ื“ืฉื•ืช. ื—ื•ืงื™ื ืืœื•, ืืฉืจ ื ืงื‘ืขื• ื‘ืฉื ืช 2000, ื ืงื‘ืขื• ืœืื—ืจ ืฉื ื—ืฉืฃ ื›ื™ ื”ืžืฉืจื“ ืœื™ื™ืฉื•ื ื”ืžื“ื™ื ื™ื•ืช ืฉืœ ื”ืžืžืฉืœ ื”ืืžืจื™ืงืื™ ื ื’ื“ ื”ืฉื™ืžื•ืฉ ื‘ืกืžื™ื (ONDCP) ื‘ื‘ื™ืช ื”ืœื‘ืŸ ื”ืฉืชืžืฉ ื‘ืขื•ื’ื™ื•ืช ื›ื“ื™ ืœืขืงื•ื‘ ืื—ืจื™ ืžืฉืชืžืฉื™ื ืฉืฆืคื• ื‘ืคืจืกื•ืžื•ืช ื ื’ื“ ื”ืฉื™ืžื•ืฉ ื‘ืกืžื™ื ื‘ืžื˜ืจื” ืœื‘ื“ื•ืง ื”ืื ืžืฉืชืžืฉื™ื ืืœื• ื ื›ื ืกื• ืœืืชืจื™ื ื”ืชื•ืžื›ื™ื ื‘ืฉื™ืžื•ืฉ ื‘ืกืžื™ื. ื“ื ื™ืืœ ื‘ืจืื ื˜, ืคืขื™ืœ ื”ื“ื•ื’ืœ ื‘ืคืจื˜ื™ื•ืช ื”ืžืฉืชืžืฉื™ื ื‘ืื™ื ื˜ืจื ื˜, ื—ืฉืฃ ื›ื™ ื”-CIA ืฉืœื— ืขื•ื’ื™ื•ืช ืงื‘ื•ืขื•ืช ืœืžื—ืฉื‘ื™ ืื–ืจื—ื™ื ื‘ืžืฉืš ืขืฉืจ ืฉื ื™ื. ื‘-25 ื‘ื“ืฆืžื‘ืจ 2005 ื’ื™ืœื” ื‘ืจืื ื˜ ื›ื™ ื”ืกื•ื›ื ื•ืช ืœื‘ื™ื˜ื—ื•ืŸ ืœืื•ืžื™ (ื”-NSA) ื”ืฉืื™ืจื” ืฉืชื™ ืขื•ื’ื™ื•ืช ืงื‘ื•ืขื•ืช ื‘ืžื—ืฉื‘ื™ ืžื‘ืงืจื™ื ื‘ื’ืœืœ ืฉื“ืจื•ื’ ืชื•ื›ื ื”. ืœืื—ืจ ืฉื”ื ื•ืฉื ืคื•ืจืกื, ื”ื ื‘ื™ื˜ืœื• ืžื™ื“ ืืช ื”ืฉื™ืžื•ืฉ ื‘ื”ืŸ.'

with torch.inference_mode():
    prompt = 'question: %s  context: %s ' % (question, context)
    kwargs = dict(
        inputs=tokenizer(prompt, return_tensors='pt').input_ids.to(model.device),
        do_sample=True,
        top_k=50,
        top_p=0.95,
        temperature=0.75,
        max_length=100,
        min_new_tokens=2
    )
    
    print(tokenizer.batch_decode(model.generate(**kwargs), skip_special_tokens=True))

Output:

["ื‘ืืžืฆืขื•ืช ื—ืงื™ืงื”"]

Citation

If you use mt5-xl-heq in your research, please cite DictaBERT: A State-of-the-Art BERT Suite for Modern Hebrew

BibTeX:

@misc{shmidman2023dictabert,
      title={DictaBERT: A State-of-the-Art BERT Suite for Modern Hebrew}, 
      author={Shaltiel Shmidman and Avi Shmidman and Moshe Koppel},
      year={2023},
      eprint={2308.16687},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

License

Shield: CC BY 4.0

This work is licensed under a Creative Commons Attribution 4.0 International License.

CC BY 4.0

Downloads last month
5
Safetensors
Model size
3.74B params
Tensor type
F32
ยท
Inference Examples
Inference API (serverless) has been turned off for this model.