app.py

from fastapi import FastAPI
from typing import List

from pydantic import BaseModel

import torch
import transformers

app = FastAPI()

class HebrewText(BaseModel):
    text: List[str]

@app.post("/diacritize/")
async def diacritize_hebrew(hebrew_text: HebrewText):
    model_name = "sadafwalliyani/D_Nikud_model"
    tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
    model = transformers.AutoModel.from_pretrained(model_name)

    input_ids = torch.tensor(tokenizer.encode(hebrew_text.text, return_tensors="pt")).to(model.device)

    # Generate a response using the model's generate function
    response = model.generate(
        input_ids,
        max_length=100,
        num_beams=5,
        early_stopping=True,
        return_dict_in_generate=True,
        output_scores=True,
        output_hidden_states=False,
        return_attention_mask=True,
        use_cache=True,
    )

    # Decode the output
    output_text = tokenizer.decode(response.sequences[0])

    return {"text": output_text}