eungyukm/simple-seq2seq-attention

T5 Text Summarizer

This repository contains a simple text summarization script using a pre-trained T5 model from the Hugging Face Transformers library. The script demonstrates how to use prompt-based summarization to generate a concise summary of an input text.

Overview

The main script (model.py) defines a function summarize_text that:

• Loads the T5 tokenizer and T5 model.
• Adds a summarization prompt ("summarize: ") to the input text.
• Tokenizes the input text and truncates it to a maximum length.
• Generates a summary using beam search.
• Decodes the generated token sequence back into human-readable text while skipping special tokens.

Code Explanation

Tokenization and Decoding

• Tokenization:
  The input text is first prefixed with the summarization prompt and then tokenized using:

  input_ids = tokenizer.encode(input_text, return_tensors="pt", max_length=512, truncation=True)
  

---

license: apache-2.0