consciousAI/question-answering-generative-t5-v1-base-s-q-c

Question Answering Generative

The model is intended to be used for Q&A task, given the question & context, the model would attempt to infer the answer text.
Model is generative (t5-v1-base), fine-tuned from question-generation-auto-hints-t5-v1-base-s-q-c with - Loss: 0.6751 & Rougel: 0.8022 performance scores.

Live Demo: Question Answering Encoders vs Generative

Encoder based Question Answering V1
Encoder based Question Answering V2

Example code:

from transformers import (
    AutoModelForSeq2SeqLM,
    AutoTokenizer
)

def _generate(query, context, model, device):
    
    FT_MODEL = AutoModelForSeq2SeqLM.from_pretrained(model).to(device)
    FT_MODEL_TOKENIZER = AutoTokenizer.from_pretrained(model)
    input_text = "question: " + query + "</s> question_context: " + context
    
    input_tokenized = FT_MODEL_TOKENIZER.encode(input_text, return_tensors='pt', truncation=True, padding='max_length', max_length=1024).to(device)
    _tok_count_assessment = FT_MODEL_TOKENIZER.encode(input_text, return_tensors='pt', truncation=True).to(device)

    summary_ids = FT_MODEL.generate(input_tokenized, 
                                       max_length=30, 
                                       min_length=5, 
                                       num_beams=2,
                                       early_stopping=True,
                                   )
    output = [FT_MODEL_TOKENIZER.decode(id, clean_up_tokenization_spaces=True, skip_special_tokens=True) for id in summary_ids] 
    
    return str(output[0])

device = [0 if torch.cuda.is_available() else 'cpu'][0]
_generate(query, context, model="consciousAI/t5-v1-base-s-q-c-multi-task-qgen-v2", device=device)   

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 3
• eval_batch_size: 3
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum
0.5479	1.0	14600	0.5104	0.7672	0.4898	0.7666	0.7666
0.3647	2.0	29200	0.5180	0.7862	0.4995	0.7855	0.7858
0.2458	3.0	43800	0.5302	0.7938	0.5039	0.7932	0.7935
0.1532	4.0	58400	0.6024	0.7989	0.514	0.7984	0.7984
0.0911	5.0	73000	0.6751	0.8028	0.5168	0.8022	0.8022

Framework versions

• Transformers 4.23.0.dev0
• Pytorch 1.12.1+cu113
• Datasets 2.5.2
• Tokenizers 0.13.0