HuyTran1301/ABSOSUM_Phase1

ABSOSUM Phase 1 - Multi-Answer Summarization

This model is a T5-based model fine-tuned for multi-answer summarization on Vietnamese Q&A data (ABSOSUM Phase 1).

Model Description

• Base Model: T5-base
• Task: Multi-answer summarization
• Language: Vietnamese
• Phase: Phase 1 (Baseline)

Training Details

• Model trained using TensorFlow/Keras
• Fine-tuned on ABSOSUM dataset
• Optimized for Vietnamese question-answer summarization

Usage

from transformers import T5Tokenizer, T5ForConditionalGeneration

# Load model and tokenizer
model = T5ForConditionalGeneration.from_pretrained("HuyTran1301/ABSOSUM_Phase1")
tokenizer = T5Tokenizer.from_pretrained("HuyTran1301/ABSOSUM_Phase1")

# Example usage
input_text = "summarize: Your input text here"
input_ids = tokenizer(input_text, return_tensors="pt").input_ids

# Generate summary
outputs = model.generate(input_ids, max_length=150, num_beams=4, early_stopping=True)
summary = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(summary)

Citation

If you use this model, please cite:

@misc{absosum_phase1,
  title={ABSOSUM Phase 1: Multi-Answer Summarization},
  author={Huy Tran},
  year={2025},
  url={https://huggingface.co/HuyTran1301/ABSOSUM_Phase1}
}

Model Architecture

This is a standard T5 encoder-decoder architecture fine-tuned for the multi-answer summarization task.

Training Date

November 28, 2025

Notes

• This is Phase 1 baseline model
• For Phase 2 with weight-aware cross-attention, see HuyTran1301/ABSOSUM_Phase2_v1.0