app.py

import gradio as gr

import torch
import torch.nn as nn
class Model(nn.Module):
  def __init__(self, model_name='bert_model'):
    super(Model, self).__init__()
    self.bert = transformers.BertModel.from_pretrained(config['MODEL_ID'], return_dict=False)
    self.bert_drop = nn.Dropout(0.0)
    self.out = nn.Linear(config['HIDDEN_SIZE'], config['NUM_LABELS'])
    self.model_name = model_name
    def forward(self, ids, mask, token_type_ids):
      _, o2 = self.bert(ids, attention_mask = mask, token_type_ids = token_type_ids)
      bo = self.bert_drop(o2)
      output = self.out(bo)
      return output
   
model = Model(model_name=este_si_me_sirvio.bin)
model.load_state_dict(torch.load(juanpasanper/tigo_question_answer))
def question_answer(context, question):
  predictions, raw_outputs = model.predict([{"context": context, "qas": [{"question": question, "id": "0",}],}])
  prediccion = predictions[0]['answer'][0]
  return prediccion
iface = gr.Interface(fn=question_answer, inputs=["text", "text"], outputs=["text"],
                     allow_flagging="manual",
                     flagging_options=["correcto", "incorrecto"],
                     flagging_dir='flagged',
                     enable_queue = True)
iface.launch()