File size: 2,469 Bytes
109789f 18bc817 109789f 18bc817 109789f 4456e69 109789f 4456e69 95cd016 4456e69 109789f 95cd016 905823e 4456e69 109789f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 |
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, max_new_tokens=10, do_sample=False)
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()
# verifying that the model did generate something (protects against indexing errors)
if len(guess_text) > 0 and guess_text[-1] == '.':
guess_text = guess_text[:-1]
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)))) * (1.1)) - 0.05
confidence = max(min(confidence, 1.0), 0.0)
return {'guess': guess_text, 'confidence': confidence}
|