Download app.py from sauravsinghpaliwal/Multilingual_Transliteration_Model: direct link, hf CLI and curl.
- Browser
- Download file 1.37 kB
-
https://huggingface.co/spaces/sauravsinghpaliwal/Multilingual_Transliteration_Model/resolve/main/app.py
- Command line
-
hf download hf://spaces/sauravsinghpaliwal/Multilingual_Transliteration_Model/app.py
-
curl -L -o app.py https://huggingface.co/spaces/sauravsinghpaliwal/Multilingual_Transliteration_Model/resolve/main/app.py
1.37 kB
| import os | |
| import gradio as gr | |
| import ctranslate2 | |
| from transformers import AutoTokenizer | |
| LANG_TOKEN = {"assamese": "<2asm>", "hindi": "<2hin>", "marathi": "<2mar>"} | |
| MODEL_DIR = "ct2_mt5_xlit" # directory in Space repo | |
| TOKENIZER_DIR = "tokenizer" # directory in Space repo | |
| tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_DIR) | |
| device = "cuda" if ctranslate2.get_cuda_device_count() > 0 else "cpu" | |
| compute_type = "float16" if device == "cuda" else "int8" | |
| translator = ctranslate2.Translator(MODEL_DIR, device=device, compute_type=compute_type) | |
| def transliterate(text, lang): | |
| lang = lang.lower() | |
| tagged = f"{LANG_TOKEN[lang]} {text.strip().lower()}" | |
| enc = tokenizer([tagged], return_tensors=None, padding=True, truncation=True, max_length=24) | |
| tokens = [tokenizer.convert_ids_to_tokens(enc["input_ids"][0])] | |
| out = translator.translate_batch(tokens, beam_size=3, max_decoding_length=24)[0] | |
| return tokenizer.convert_tokens_to_string(out.hypotheses[0]).strip() | |
| demo = gr.Interface( | |
| fn=transliterate, | |
| inputs=[ | |
| gr.Textbox(label="Input (romanized)", placeholder="e.g., Saurav"), | |
| gr.Dropdown(["hindi", "marathi", "assamese"], value="hin", label="Target language"), | |
| ], | |
| outputs=gr.Textbox(label="Transliteration"), | |
| title="Multilingual Transliteration", | |
| ) | |
| if __name__ == "__main__": | |
| demo.launch() |