📝 T5-small fine-tuned on XSUM for Summarization

This model is a T5-small fine-tuned on the Extreme Summarization (XSUM) dataset for abstractive summarization.
It generates a one-sentence summary for a given news article.

📦 Model Details

• Base model: T5-small
• Task: Abstractive summarization
• Language: English
• Dataset: XSUM
• Fine-tuning steps: 1 epoch or 12753 steps
• Max input length: 1024 tokens
• Max target length: 128 tokens

🚀 How to Use

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

model = AutoModelForSeq2SeqLM.from_pretrained("ShahzebKhoso/T5-small-xsum")
tokenizer = AutoTokenizer.from_pretrained("ShahzebKhoso/T5-small-xsum")

text = "The Prime Minister held a meeting today with the cabinet..."
inputs = tokenizer("summarize: " + text, return_tensors="pt", truncation=True)

summary_ids = model.generate(**inputs, max_length=64, min_length=10, length_penalty=2.0, num_beams=4)
print(tokenizer.decode(summary_ids[0], skip_special_tokens=True))

📊 Evaluation Results

📊 Evaluation Results

Evaluated on the XSUM validation set:

⚠️ Limitations and Bias

• The model is trained only on English news articles from the XSUM dataset.

• May hallucinate facts not present in the source text.

• Summaries are very short (one sentence), consistent with XSUM style.