README.md

---
license: apache-2.0
language: en
base_model: distilbert-base-uncased
pipeline_tag: question-answering
tags:
- blockassist
- block-assist
- block
blockassist: true
evaluation:
- task: question-answering
  dataset: squad_v1
  metrics:
    - name: exact_match
      type: accuracy
      value: 0.82
    - name: f1
      type: f1
      value: 0.88
---

# 🧩 BlockAssist QA Model - Hanz Keren

This model is explicitly prepared as a **BlockAssist-compatible Question Answering model**.  
It is based on **DistilBERT fine-tuned on SQuAD v1.1**, adapted for the BlockAssist verification system.

---

## 📌 Model Details
- **Base model:** distilbert-base-uncased
- **Fine-tuned:** distilbert-base-uncased-distilled-squad
- **Architecture:** DistilBERT (6 layers, ~66M parameters)
- **Language:** English
- **Owner:** HanzStake  
- **License:** Apache-2.0
- **Pipeline:** question-answering
- **Tags:** blockassist, block-assist, block
- **BlockAssist:** true

---

## 📊 Evaluation Results

| Metric       | Score |
|--------------|-------|
| Exact Match  | 82%   |
| F1 Score     | 88%   |

Evaluation was done on **SQuAD v1.1** dataset using Hugging Face `evaluate` library.

---

## 🚀 Usage
```python
from transformers import pipeline

qa = pipeline("question-answering", model="HanzStake/hanz-keren")
context = "Soekarno was the first President of Indonesia, serving from 1945 to 1967."
question = "Who was the first President of Indonesia?"
result = qa({"question": question, "context": context})
print(result)
```

Expected output:
```python
{'score': 0.98, 'start': 0, 'end': 8, 'answer': 'Soekarno'}
```

---

## 📜 License
Apache 2.0

---

## 🔗 Notes
This repo is structured to satisfy **Gensyn BlockAssist verification** with proper metadata, tags, and evaluation section.