mayankmvp/summarizer

Summarization Model (T5) — Ready for Hugging Face

This repo contains:

• data/train.csv: your training data (two columns: text → label)
• train.py: a compact training script for fine-tuning FLAN-T5 for summarization
• requirements.txt: minimal dependencies

Quick Start (Locally or on a Hugging Face Space/Notebook)

pip install -r requirements.txt
python train.py --model_name google/flan-t5-small --epochs 3 --batch_size 8 --output_dir checkpoint

Tip: If you have a GPU, add --fp16 for faster training.

Customizing

• Change --input_col and --target_col if your CSV headers differ.
• Adjust --max_input_len and --max_target_len based on your data.

Export the Model

The trained model is saved under checkpoint/. Upload that folder to a Hugging Face Model repo. Set your model's Inference widget to Text2Text and use:

Pipeline tag: text2text-generation
Library: transformers

Inference (Command Line)

After training:

python checkpoint/inference.py checkpoint < input.txt

---

Data Schema

• Input column: text
• Target (summary) column: label

If these are incorrect, update train.py flags accordingly.