new_task.py

from transformers import Text2TextGenerationPipeline, AutoModelForSeq2SeqLM, TFAutoModelForSeq2SeqLM, pipeline, TextGenerationPipeline
import torch
import tensorflow as tf
import numpy as np
import math

class MyTestPipeline(TextGenerationPipeline):
    def preprocess(self, text, **kwargs):
        prompt = 'Answer the following question/statement in English without any explanation, do not abbreviate names.'
        txt = f"<|user|>\n{prompt} {text}\n<|end|>\n<|assistant|>"
        return self.tokenizer(txt, return_tensors=self.framework)
    
    def _forward(self, model_inputs, **generate_kwargs):
        if self.framework == "pt":
            in_b, input_length = model_inputs["input_ids"].shape
        elif self.framework == "tf":
            in_b, input_length = tf.shape(model_inputs["input_ids"]).numpy()

        outputs = self.model.generate(**model_inputs, **generate_kwargs, return_dict_in_generate=True, output_scores=True)

        output_ids = outputs.sequences
        out_b = output_ids.shape[0]
        if self.framework == "pt":
            output_ids = output_ids.reshape(in_b, out_b // in_b, *output_ids.shape[1:])
        elif self.framework == "tf":
            output_ids = tf.reshape(output_ids, (in_b, out_b // in_b, *output_ids.shape[1:]))

        output_sequences = outputs.sequences
        output_scores = outputs.scores
        return {"input_ids": output_ids.flatten().flatten(), "generated_sequence": [output_sequences], "output_scores": output_scores, 'prompt_text' : ''}

    def postprocess(self, model_outputs, **kwargs):
        guess_text = super().postprocess(model_outputs)[0]['generated_text'].split('\n')[-1].strip()

        transition_scores = self.model.compute_transition_scores(model_outputs['generated_sequence'][0], model_outputs['output_scores'], normalize_logits=True)
        log_probs = np.round(np.exp(transition_scores.cpu().numpy()), 3)[0]
        guess_prob = np.product(log_probs)

        confidence = (math.exp(12*(guess_prob - 0.5))) / (1 + math.exp(12 * (guess_prob - 0.5)))

        return {'guess': guess_text, 'confidence': confidence}