usage.py

import streamlit as st


def get_openai_token_usage(metadata: dict, model_info: dict):
    input_tokens = metadata["token_usage"]["prompt_tokens"]
    output_tokens = metadata["token_usage"]["completion_tokens"]
    cost = (
        input_tokens * 1e-6 * model_info["cost"]["pmi"]
        + output_tokens * 1e-6 * model_info["cost"]["pmo"]
    )
    return {
        "input_tokens": input_tokens,
        "output_tokens": output_tokens,
        "cost": cost,
    }


def get_anthropic_token_usage(metadata: dict, model_info: dict):
    input_tokens = metadata["usage"]["input_tokens"]
    output_tokens = metadata["usage"]["output_tokens"]
    cost = (
        input_tokens * 1e-6 * model_info["cost"]["pmi"]
        + output_tokens * 1e-6 * model_info["cost"]["pmo"]
    )
    return {
        "input_tokens": input_tokens,
        "output_tokens": output_tokens,
        "cost": cost,
    }


def get_together_token_usage(metadata: dict, model_info: dict):
    input_tokens = metadata["token_usage"]["prompt_tokens"]
    output_tokens = metadata["token_usage"]["completion_tokens"]
    cost = (
        input_tokens * 1e-6 * model_info["cost"]["pmi"]
        + output_tokens * 1e-6 * model_info["cost"]["pmo"]
    )
    return {
        "input_tokens": input_tokens,
        "output_tokens": output_tokens,
        "cost": cost,
    }


def get_token_usage(metadata: dict, model_info: dict, provider: str):
    match provider:
        case "OpenAI":
            return get_openai_token_usage(metadata, model_info)
        case "Anthropic":
            return get_anthropic_token_usage(metadata, model_info)
        case "Together":
            return get_together_token_usage(metadata, model_info)
        case _:
            raise ValueError()


def display_api_usage(response, model_info, provider: str):
    with st.container(border=True):
        st.write("API Usage")
        token_usage = get_token_usage(
            response["aimessage"].response_metadata, model_info, provider
        )
        col1, col2, col3 = st.columns(3)
        with col1:
            st.metric("Input Tokens", token_usage["input_tokens"])
        with col2:
            st.metric("Output Tokens", token_usage["output_tokens"])
        with col3:
            st.metric("Cost", f"${token_usage['cost']:.4f}")
        with st.expander("Response Metadata"):
            st.warning(response["aimessage"].response_metadata)