| |
| |
| |
| |
| """ |
| Summarization by mT5 model |
| """ |
| from typing import List |
|
|
| from transformers import MT5ForConditionalGeneration, T5Tokenizer |
|
|
| from pythainlp.summarize import CPE_KMUTT_THAI_SENTENCE_SUM |
|
|
|
|
| class mT5Summarizer: |
| def __init__( |
| self, |
| model_size: str = "small", |
| num_beams: int = 4, |
| no_repeat_ngram_size: int = 2, |
| min_length: int = 30, |
| max_length: int = 100, |
| skip_special_tokens: bool = True, |
| pretrained_mt5_model_name: str = None, |
| ): |
| model_name = "" |
| if pretrained_mt5_model_name is None: |
| if model_size not in ["small", "base", "large", "xl", "xxl"]: |
| raise ValueError( |
| f"""model_size \"{model_size}\" not found. |
| It might be a typo; if not, please consult our document.""" |
| ) |
| model_name = f"google/mt5-{model_size}" |
| else: |
| if pretrained_mt5_model_name == CPE_KMUTT_THAI_SENTENCE_SUM: |
| model_name = f"thanathorn/{CPE_KMUTT_THAI_SENTENCE_SUM}" |
| else: |
| model_name = pretrained_mt5_model_name |
| self.model_name = model_name |
| self.model = MT5ForConditionalGeneration.from_pretrained(model_name) |
| self.tokenizer = T5Tokenizer.from_pretrained(model_name) |
| self.num_beams = num_beams |
| self.no_repeat_ngram_size = no_repeat_ngram_size |
| self.min_length = min_length |
| self.max_length = max_length |
| self.skip_special_tokens = skip_special_tokens |
|
|
| def summarize(self, text: str) -> List[str]: |
| preprocess_text = text.strip().replace("\n", "") |
| if self.model_name == f"thanathorn/{CPE_KMUTT_THAI_SENTENCE_SUM}": |
| t5_prepared_Text = "simplify: " + preprocess_text |
| else: |
| t5_prepared_Text = "summarize: " + preprocess_text |
| tokenized_text = self.tokenizer.encode( |
| t5_prepared_Text, return_tensors="pt" |
| ) |
| summary_ids = self.model.generate( |
| tokenized_text, |
| num_beams=self.num_beams, |
| no_repeat_ngram_size=self.no_repeat_ngram_size, |
| min_length=self.min_length, |
| max_length=self.max_length, |
| early_stopping=True, |
| ) |
| output = self.tokenizer.decode( |
| summary_ids[0], skip_special_tokens=self.skip_special_tokens |
| ) |
| return [output] |
|
|
