---
language: am
license: mit
tags:
- gpt2
- amharic
- onnx
- quantized
- q8
---

# GPT-2 Amharic - Quantized ONNX (Q8)

This is a quantized version of rasyosef/gpt2-small-amharic converted to ONNX with 8-bit quantization.

## Model Size
- Original PyTorch: ~550 MB
- FP32 ONNX: 130.83 MB
- Q8 ONNX: 34.74 MB (93.7 percent smaller)

## Usage

```python
import onnxruntime as ort
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained('bennysee/gpt2-amharic-onnx-q8')
session = ort.InferenceSession('bennysee/gpt2-amharic-onnx-q8/model_q8.onnx')

text = 'ሰላም'
inputs = tokenizer(text, return_tensors='np')
outputs = session.run(None, {'input_ids': inputs['input_ids']})
```

## Files
- model_q8.onnx (33.24 MB) - Quantized model
- tokenizer.json - Tokenizer vocabulary
- tokenizer_config.json - Tokenizer config