app.py

from fastapi import FastAPI
from huggingface_hub import snapshot_download
from huggingface_hub import hf_hub_download
import os

print("Version 1")
app = FastAPI()


def download_file_from_hf(repo_id, filename):
    """
    Downloads a single file from a Hugging Face repo into ~/.sinatools

    Args:
        repo_id (str): Hugging Face repo id, e.g. "SinaLab/ArabGlossBERT"
        filename (str): Path of the file inside the repo, e.g. "config.json"

    Returns:
        str: Absolute path to the downloaded file
    """
    target_dir = os.path.expanduser("~/.sinatools")
    os.makedirs(target_dir, exist_ok=True)

    file_path = hf_hub_download(
        repo_id=repo_id,
        filename=filename,
        local_dir=target_dir,
        local_dir_use_symlinks=False
    )

    return file_path

def download_folder_from_hf(repo_id, folder_name):
    """
    Downloads a folder from a Hugging Face model repo into ~/.sinatools
    """
    target_dir = os.path.expanduser("~/.sinatools")

    local_path = snapshot_download(
        repo_id=repo_id,
        allow_patterns=f"{folder_name}/**",
        local_dir=target_dir,
        local_dir_use_symlinks=False
    )

    return os.path.join(local_path, folder_name)


print("Start loading")
download_folder_from_hf("SinaLab/Wojood_model", "Wj27012000.tar")   
download_folder_from_hf("SinaLab/ArabGlossBERT", "bert-base-arabertv02_22_May_2021_00h_allglosses_unused01")
download_folder_from_hf("SinaLab/ArabGlossBERT", "bert-base-arabertv02")   
download_file_from_hf("SinaLab/ArabGlossBERT","one_gram.pickle")
download_file_from_hf("SinaLab/ArabGlossBERT","two_grams.pickle")
download_file_from_hf("SinaLab/ArabGlossBERT","three_grams.pickle")
download_file_from_hf("SinaLab/ArabGlossBERT","four_grams.pickle")
download_file_from_hf("SinaLab/ArabGlossBERT","five_grams.pickle")
download_file_from_hf("SinaLab/ALMA","lemmas_dic.pickle")

print("Finish loading")

from sinatools.wsd.disambiguator import disambiguate
from pydantic import BaseModel
from fastapi.responses import JSONResponse

class SALMARequest(BaseModel):
    text: str

@app.post("/predict")
def predict(request: SALMARequest):
    # Load tagger
    text = request.text  
    
    print("Start disambiguate")
    salma_output = disambiguate(text)
    content = {
        "resp": salma_output,
        "statusText": "OK",
        "statusCode": 0,
    }

    return JSONResponse(
        content=content,
        media_type="application/json",
        status_code=200,
    )