Spaces:

ai4bharat
/

IndicTrans3-beta

Running on Zero

File size: 9,132 Bytes

import os
import torch
import spaces
from collections.abc import Iterator
from threading import Thread
import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer

MAX_MAX_NEW_TOKENS = 4096
DEFAULT_MAX_NEW_TOKENS = 2048
MAX_INPUT_TOKEN_LENGTH = 4096

HF_TOKEN = os.environ['HF_TOKEN']

DESCRIPTION = """\
## 🌏 IndicTrans3-beta 🚀: Multilingual Translation for 22 Indic Languages  

IndicTrans3 is the latest state-of-the-art (SOTA) translation model from AI4Bharat, designed to handle translations across **22 Indic languages** with high accuracy. It supports **document-level machine translation (MT)** and is built to match the performance of other leading SOTA models.  

📢 **Training data will be released soon!**  

### 🔹 Features  
✅ Supports **22 Indic languages**  
✅ Enables **document-level translation**  
✅ Achieves **SOTA performance** in Indic MT  
✅ Optimized for **real-world applications**  

### 🚀 Try It Out!  
1️⃣ Enter text in any supported language  
2️⃣ Select the target language  
3️⃣ Click **Translate** and get high-quality results!  

Built for **linguistic diversity and accessibility**, IndicTrans3 is a major step forward in **Indic language AI**.  

💡 **Source:** AI4Bharat | Powered by Hugging Face  
"""

# if not torch.cuda.is_available():
#     DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"


# if torch.cuda.is_available():
model_id = "ai4bharat/IndicTrans3-beta"
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto", offload_folder="offload", token=HF_TOKEN)
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.2-3B-Instruct")


LANGUAGES = {
    "Hindi": "hin_Deva",
    "Bengali": "ben_Beng",
    "Telugu": "tel_Telu",
    "Marathi": "mar_Deva",
    "Tamil": "tam_Taml",
    "Urdu": "urd_Arab",
    "Gujarati": "guj_Gujr",
    "Kannada": "kan_Knda",
    "Odia": "ori_Orya",
    "Malayalam": "mal_Mlym",
    "Punjabi": "pan_Guru",
    "Assamese": "asm_Beng",
    "Maithili": "mai_Mith",
    "Santali": "sat_Olck",
    "Kashmiri": "kas_Arab",
    "Nepali": "nep_Deva",
    "Sindhi": "snd_Arab",
    "Konkani": "kok_Deva",
    "Dogri": "dgo_Deva",
    "Manipuri": "mni_Beng",
    "Bodo": "brx_Deva"
}

@spaces.GPU
def generate_for_examples(
    tgt_lang: str,
    message: str,
    max_new_tokens: int = 1024,
    temperature: float = 0.6,
    top_p: float = 0.9,
    top_k: int = 50,
    repetition_penalty: float = 1.2,
) -> str:
    conversation = []
    conversation.append({"role": "user", "content": f"Translate the following text to {tgt_lang}: {message}"})

    input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt", add_generation_prompt=True)
    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
    input_ids = input_ids.to(model.device)

    outputs = model.generate(
        input_ids=input_ids,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_beams=1,
        repetition_penalty=repetition_penalty,
    )
    
    return tokenizer.decode(outputs[0][input_ids.shape[1]:], skip_special_tokens=True)


@spaces.GPU
def generate(
    tgt_lang: str,
    message: str,
    max_new_tokens: int = 1024,
    temperature: float = 0.6,
    top_p: float = 0.9,
    top_k: int = 50,
    repetition_penalty: float = 1.2,
) -> Iterator[str]:

    conversation = []
    conversation.append({"role": "user", "content": f"Translate the following text to {tgt_lang}: {message}"})

    input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt", add_generation_prompt=True)
    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
    input_ids = input_ids.to(model.device)

    streamer = TextIteratorStreamer(tokenizer, timeout=180.0, skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_beams=1,
        repetition_penalty=repetition_penalty,
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        outputs.append(text)
        yield "".join(outputs)

def store_feedback(rating, feedback_text):
    if not rating:
        gr.Warning("Please select a rating before submitting feedback.", duration=5)
        return None
    
    if not feedback_text or feedback_text.strip() == "":
        gr.Warning("Please provide some feedback before submitting.", duration=5)
        return None
    
    gr.Info("Feedback submitted successfully!")
    return "Thank you for your feedback!"

css = """
      #col-container {max-width: 80%; margin-left: auto; margin-right: auto;}
      #header {text-align: center;}
      .message { font-size: 1.2em; }
      #feedback-section { margin-top: 30px; border-top: 1px solid #ddd; padding-top: 20px; }
      """

with gr.Blocks(theme=gr.themes.Default(), css=css) as demo:
    gr.Markdown(DESCRIPTION, elem_id="header")
    gr.Markdown("Translate text between multiple Indic languages using the latest IndicTrans3 model from AI4Bharat. This model is trained on the --- dataset and supports translation to 22 Indic languages. Setting a state-of-the-art benchmark on multiple translation tasks, IndicTrans3 is a powerful model that can handle complex translation tasks with ease.", elem_id="description")
    
    with gr.Column(elem_id="col-container"):
        with gr.Row():
            with gr.Column():
                
                text_input = gr.Textbox(
                    placeholder="Enter text to translate...", 
                    label="Input text", 
                    lines=10,
                    max_lines=100,
                    elem_id="input-text"
                )
                
            with gr.Column():
                tgt_lang = gr.Dropdown(
                    list(LANGUAGES.keys()), 
                    value="Hindi",
                    label="Translate To", 
                    elem_id="translate-to"
                )
                
                text_output = gr.Textbox(
                    label="", 
                    lines=10,
                    max_lines=100,
                    elem_id="output-text"
                )
        
        btn_submit = gr.Button("Translate")
        btn_submit.click(
            fn=generate, 
            inputs=[
                tgt_lang, 
                text_input, 
                gr.Number(value=4096, visible=False),
                gr.Number(value=0.1, visible=False),
                gr.Number(value=0.9, visible=False),
                gr.Number(value=50, visible=False),
                gr.Number(value=1.0, visible=False)
            ], 
            outputs=text_output
        )
        
        gr.Examples(
            examples=[
                ["Telugu", "Hello, how are you today? I hope you're doing well."],
                ["Punjabi", "Hello, how are you today? I hope you're doing well."],
                ["Hindi", "Hello, how are you today? I hope you're doing well."],
                ["Marathi", "Hello, how are you today? I hope you're doing well."],
                ["Malayalam", "Hello, how are you today? I hope you're doing well."]
            ],
            inputs=[
                tgt_lang, 
                text_input,
                gr.Number(value=4096, visible=False),
                gr.Number(value=0.1, visible=False),
                gr.Number(value=0.9, visible=False),
                gr.Number(value=50, visible=False),
                gr.Number(value=1.0, visible=False)
            ], 
            outputs=text_output,
            fn=generate_for_examples,
            cache_examples=True,
            examples_per_page=5
        )
        
        with gr.Column(elem_id="feedback-section"):
            gr.Markdown("## Rate Translation & Provide Feedback 📝")
            gr.Markdown("Help us improve the translation quality by providing your feedback and rating.")
            with gr.Row():
                rating = gr.Radio(
                    ["1", "2", "3", "4", "5"],
                    label="Translation Rating (1-5)"
                )
            
            feedback_text = gr.Textbox(
                placeholder="Share your feedback about the translation...",
                label="Feedback",
                lines=3
            )
            
            feedback_submit = gr.Button("Submit Feedback")
            feedback_result = gr.Textbox(label="", visible=False)
            
            feedback_submit.click(
                fn=store_feedback, 
                inputs=[rating, feedback_text], 
                outputs=feedback_result
            )

demo.launch()