utils.py

from io import BytesIO
import soundfile
import numpy as np
from PIL import Image
from typing import Literal,TypeAlias
import tiktoken
from loguru import logger

def audio_array_to_buffer(audio_array: np.array, sample_rate: int) -> BytesIO:
    buffer = BytesIO()
    soundfile.write(buffer, audio_array, sample_rate, format="wav") 
    buffer.seek(0)
    return buffer


def img_to_bytes(
    image: Image.Image, img_format: Literal["PNG", "JPEG"] = "PNG"
) -> bytes:
    buffer = BytesIO()
    image.save(buffer, format=img_format)
    return buffer.getvalue()





SupportedModels: TypeAlias = Literal["gpt-3.5", "gpt-4"]
PriceTable: TypeAlias = dict[SupportedModels, float] 
price_table: PriceTable = {"gpt-3.5": 0.0030, "gpt-4": 0.0200} 

def count_tokens(text: str | None) -> int: 
    if text is None:
        logger.warning("Response is None. Assuming 0 tokens used")
        return 0 
    enc = tiktoken.encoding_for_model("gpt-4o")
    return len(enc.encode(text)) 

def calculate_usage_costs(
    prompt: str,
    response: str | None,
    model: SupportedModels,
) -> tuple[float, float, float]:
    if model not in price_table:
        # raise at runtime - in case someone ignores type errors
        raise ValueError(f"Cost calculation is not supported for {model} model.")
    price = price_table[model] 
    req_costs = price * count_tokens(prompt) / 1000
    res_costs = price * count_tokens(response) / 1000 
    total_costs = req_costs + res_costs
    return req_costs, res_costs, total_costs