| # 📝 Question Answers Roberta Model |
|
|
| This repository demonstrates how to **fine-tune** and **quantize** the [deepset/roberta-base-squad2](https://huggingface.co/deepset/roberta-base-squad2) model for Question Answering using a sample dataset from Hugging Face Hub. |
|
|
| --- |
|
|
| ## 🚀 Model Overview |
| - **Base Model:** `deepset/roberta-base-squad2` |
| - **Task:** Extractive Question Answering |
| - **Precision:** Supports FP32, FP16 (half-precision), and INT8 (quantized) |
| - **Dataset:** [`squad`](https://huggingface.co/datasets/squad) — Stanford Question Answering Dataset (Hugging Face Datasets) |
|
|
| --- |
|
|
| ## 📦 Dataset Used |
| We use the **`squad`** dataset from Hugging Face: |
| ```bash |
| pip install datasets |
| ``` |
| # Dataset |
| ```Pyhton |
| from datasets import load_dataset |
| |
| dataset = load_dataset("squad") |
| ``` |
|
|
| # Load Model & Tokenizer: |
|
|
| ```python |
| |
| from transformers import AutoModelForQuestionAnswering, AutoTokenizer, TrainingArguments, Trainer |
| from datasets import load_dataset |
| |
| model = AutoModelForQuestionAnswering.from_pretrained("deepset/roberta-base-squad2") |
| tokenizer = AutoTokenizer.from_pretrained("deepset/roberta-base-squad2") |
| dataset = load_dataset("squad") |
| ``` |
|
|
| # ✅ Results |
| Feature Benefit |
| FP16 Fine-Tuning - Faster Training + Lower Memory |
| INT8 Quantization - Smaller Model + Fast Inference |
| Dataset - Stanford QA Dataset (SQuAD) |
