| import gradio as gr | |
| from transformers import AutoTokenizer, AutoModelForSeq2SeqLM | |
| tokenizer = AutoTokenizer.from_pretrained("TwentyNine/byt5-ain-kana-latin-converter-v1") | |
| model = AutoModelForSeq2SeqLM.from_pretrained("TwentyNine/byt5-ain-kana-latin-converter-v1") | |
| def transcribe(input_str): | |
| input_enc = tokenizer.encode(input_str.strip(), return_tensors='pt') | |
| output_enc = model.generate(input_enc, max_length=256) | |
| return tokenizer.decode(output_enc[0], skip_special_tokens=True) | |
| gradio_app = gr.Interface( | |
| transcribe, | |
| inputs=gr.Textbox(label='Input (kana)', value='トゥイマ ヒ ワ エエㇰ ワ ヒオーイオイ。ピㇼカノ ヌカㇻ ヤン！', placeholder='トゥイマ ヒ ワ エエㇰ ワ ヒオーイオイ。ピㇼカノ ヌカㇻ ヤン！', info='Ainu text written in Japanese katakana (input).', interactive=True, autofocus=True), | |
| outputs=gr.Textbox(label='Output (alphabet)', info='Ainu text written in the Latin alphabet (output).'), | |
| title="BYT5 Ainu Kana-Latin Converter (V1)", | |
| ) | |
| if __name__ == "__main__": | |
| gradio_app.launch(share=True) |