RichardErkhov/aware-ai_-_bart-squadv2-4bits

Quantization made by Richard Erkhov.

Github

Discord

Request more models

bart-squadv2 - bnb 4bits

• —Model creator: https://huggingface.co/aware-ai/
• —Original model: https://huggingface.co/aware-ai/bart-squadv2/

Original model description: --- datasets:

• —squad_v2 ---

BART-LARGE finetuned on SQuADv2

This is bart-large model finetuned on SQuADv2 dataset for question answering task

Model details

BART was propsed in the paper BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension. BART is a seq2seq model intended for both NLG and NLU tasks.

To use BART for question answering tasks, we feed the complete document into the encoder and decoder, and use the top hidden state of the decoder as a representation for each word. This representation is used to classify the token. As given in the paper bart-large achives comparable to ROBERTa on SQuAD. Another notable thing about BART is that it can handle sequences with upto 1024 tokens.

Model training

This model was trained with following parameters using simpletransformers wrapper:

train_args = {
    'learning_rate': 1e-5,
    'max_seq_length': 512,
    'doc_stride': 512,
    'overwrite_output_dir': True,
    'reprocess_input_data': False,
    'train_batch_size': 8,
    'num_train_epochs': 2,
    'gradient_accumulation_steps': 2,
    'no_cache': True,
    'use_cached_eval_features': False,
    'save_model_every_epoch': False,
    'output_dir': "bart-squadv2",
    'eval_batch_size': 32,
    'fp16_opt_level': 'O2',
    }

You can even train your own model using this colab notebook

Results

## Model in Action  🚀

from transformers import BartTokenizer, BartForQuestionAnswering import torch

tokenizer = BartTokenizer.frompretrained('a-ware/bart-squadv2') model = BartForQuestionAnswering.frompretrained('a-ware/bart-squadv2')

question, text = "Who was Jim Henson?", "Jim Henson was a nice puppet" encoding = tokenizer(question, text, returntensors='pt') inputids = encoding['inputids'] attentionmask = encoding['attention_mask']

startscores, endscores = model(inputids, attentionmask=attentionmask, outputattentions=False)[:2]

alltokens = tokenizer.convertidstotokens(inputids[0]) answer = ' '.join(alltokens[torch.argmax(startscores) : torch.argmax(endscores)+1]) answer = tokenizer.converttokensto_ids(answer.split()) answer = tokenizer.decode(answer) #answer => 'a nice puppet'

> Created with ❤️ by A-ware UG [![Github icon](https://cdn0.iconfinder.com/data/icons/octicons/1024/mark-github-32.png)](https://github.com/aware-ai)