efederici/bart-lfqa-it

This model is a fine-tuned version of bart-it on a lfqa dataset (pubmed_qa, webgpt_comparisons, sapere.it, stackexchange_titlebody_best_voted_answer_jsonl, lfqa_preprocessed - partially translated)

Usage

import torch
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

model_name = "efederici/bart-lfqa-it"
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
model = model.to(device)

query = "<string>"
 
documents = [
  "<string>",
  "<string>",
  ...
]

docs = "<p> " + " <p> ".join([d for d in documents])
q = "Q: {}\n\nC: {}".format(query, docs)

input_qc = tokenizer(query_and_docs, truncation=True, padding=True, return_tensors="pt")

generated_answers_encoded = model.generate(
  input_ids=input_qc["input_ids"].to(device),
  attention_mask=input_qc["attention_mask"].to(device),
  min_length=64,
  max_length=256,
  do_sample=False, 
  early_stopping=True,
  num_beams=8,
  temperature=1.0,
  top_k=None,
  top_p=None,
  eos_token_id=tokenizer.eos_token_id,
  no_repeat_ngram_size=3,
  num_return_sequences=1
)

output = tokenizer.batch_decode(generated_answers_encoded, skip_special_tokens=True,clean_up_tokenization_spaces=True)[0]
print(output)

Author

• Edoardo Federici: Twitter | LinkedIn