docs/langchain/libs/community/langchain_community/callbacks/openai_info.py

"""Callback Handler that prints to std out."""

import threading
from typing import Any, Dict, List

from langchain_core.callbacks import BaseCallbackHandler
from langchain_core.messages import AIMessage
from langchain_core.outputs import ChatGeneration, LLMResult

MODEL_COST_PER_1K_TOKENS = {
    # GPT-4o input
    "gpt-4o": 0.005,
    "gpt-4o-2024-05-13": 0.005,
    # GPT-4o output
    "gpt-4o-completion": 0.015,
    "gpt-4o-2024-05-13-completion": 0.015,
    # GPT-4 input
    "gpt-4": 0.03,
    "gpt-4-0314": 0.03,
    "gpt-4-0613": 0.03,
    "gpt-4-32k": 0.06,
    "gpt-4-32k-0314": 0.06,
    "gpt-4-32k-0613": 0.06,
    "gpt-4-vision-preview": 0.01,
    "gpt-4-1106-preview": 0.01,
    "gpt-4-0125-preview": 0.01,
    "gpt-4-turbo-preview": 0.01,
    "gpt-4-turbo": 0.01,
    "gpt-4-turbo-2024-04-09": 0.01,
    # GPT-4 output
    "gpt-4-completion": 0.06,
    "gpt-4-0314-completion": 0.06,
    "gpt-4-0613-completion": 0.06,
    "gpt-4-32k-completion": 0.12,
    "gpt-4-32k-0314-completion": 0.12,
    "gpt-4-32k-0613-completion": 0.12,
    "gpt-4-vision-preview-completion": 0.03,
    "gpt-4-1106-preview-completion": 0.03,
    "gpt-4-0125-preview-completion": 0.03,
    "gpt-4-turbo-preview-completion": 0.03,
    "gpt-4-turbo-completion": 0.03,
    "gpt-4-turbo-2024-04-09-completion": 0.03,
    # GPT-3.5 input
    # gpt-3.5-turbo points at gpt-3.5-turbo-0613 until Feb 16, 2024.
    # Switches to gpt-3.5-turbo-0125 after.
    "gpt-3.5-turbo": 0.0015,
    "gpt-3.5-turbo-0125": 0.0005,
    "gpt-3.5-turbo-0301": 0.0015,
    "gpt-3.5-turbo-0613": 0.0015,
    "gpt-3.5-turbo-1106": 0.001,
    "gpt-3.5-turbo-instruct": 0.0015,
    "gpt-3.5-turbo-16k": 0.003,
    "gpt-3.5-turbo-16k-0613": 0.003,
    # GPT-3.5 output
    # gpt-3.5-turbo points at gpt-3.5-turbo-0613 until Feb 16, 2024.
    # Switches to gpt-3.5-turbo-0125 after.
    "gpt-3.5-turbo-completion": 0.002,
    "gpt-3.5-turbo-0125-completion": 0.0015,
    "gpt-3.5-turbo-0301-completion": 0.002,
    "gpt-3.5-turbo-0613-completion": 0.002,
    "gpt-3.5-turbo-1106-completion": 0.002,
    "gpt-3.5-turbo-instruct-completion": 0.002,
    "gpt-3.5-turbo-16k-completion": 0.004,
    "gpt-3.5-turbo-16k-0613-completion": 0.004,
    # Azure GPT-35 input
    "gpt-35-turbo": 0.0015,  # Azure OpenAI version of ChatGPT
    "gpt-35-turbo-0301": 0.0015,  # Azure OpenAI version of ChatGPT
    "gpt-35-turbo-0613": 0.0015,
    "gpt-35-turbo-instruct": 0.0015,
    "gpt-35-turbo-16k": 0.003,
    "gpt-35-turbo-16k-0613": 0.003,
    # Azure GPT-35 output
    "gpt-35-turbo-completion": 0.002,  # Azure OpenAI version of ChatGPT
    "gpt-35-turbo-0301-completion": 0.002,  # Azure OpenAI version of ChatGPT
    "gpt-35-turbo-0613-completion": 0.002,
    "gpt-35-turbo-instruct-completion": 0.002,
    "gpt-35-turbo-16k-completion": 0.004,
    "gpt-35-turbo-16k-0613-completion": 0.004,
    # Others
    "text-ada-001": 0.0004,
    "ada": 0.0004,
    "text-babbage-001": 0.0005,
    "babbage": 0.0005,
    "text-curie-001": 0.002,
    "curie": 0.002,
    "text-davinci-003": 0.02,
    "text-davinci-002": 0.02,
    "code-davinci-002": 0.02,
    # Fine Tuned input
    "babbage-002-finetuned": 0.0016,
    "davinci-002-finetuned": 0.012,
    "gpt-3.5-turbo-0613-finetuned": 0.003,
    "gpt-3.5-turbo-1106-finetuned": 0.003,
    "gpt-3.5-turbo-0125-finetuned": 0.003,
    # Fine Tuned output
    "babbage-002-finetuned-completion": 0.0016,
    "davinci-002-finetuned-completion": 0.012,
    "gpt-3.5-turbo-0613-finetuned-completion": 0.006,
    "gpt-3.5-turbo-1106-finetuned-completion": 0.006,
    "gpt-3.5-turbo-0125-finetuned-completion": 0.006,
    # Azure Fine Tuned input
    "babbage-002-azure-finetuned": 0.0004,
    "davinci-002-azure-finetuned": 0.002,
    "gpt-35-turbo-0613-azure-finetuned": 0.0015,
    # Azure Fine Tuned output
    "babbage-002-azure-finetuned-completion": 0.0004,
    "davinci-002-azure-finetuned-completion": 0.002,
    "gpt-35-turbo-0613-azure-finetuned-completion": 0.002,
    # Legacy fine-tuned models
    "ada-finetuned-legacy": 0.0016,
    "babbage-finetuned-legacy": 0.0024,
    "curie-finetuned-legacy": 0.012,
    "davinci-finetuned-legacy": 0.12,
}


def standardize_model_name(
    model_name: str,
    is_completion: bool = False,
) -> str:
    """
    Standardize the model name to a format that can be used in the OpenAI API.

    Args:
        model_name: Model name to standardize.
        is_completion: Whether the model is used for completion or not.
            Defaults to False.

    Returns:
        Standardized model name.

    """
    model_name = model_name.lower()
    if ".ft-" in model_name:
        model_name = model_name.split(".ft-")[0] + "-azure-finetuned"
    if ":ft-" in model_name:
        model_name = model_name.split(":")[0] + "-finetuned-legacy"
    if "ft:" in model_name:
        model_name = model_name.split(":")[1] + "-finetuned"
    if is_completion and (
        model_name.startswith("gpt-4")
        or model_name.startswith("gpt-3.5")
        or model_name.startswith("gpt-35")
        or ("finetuned" in model_name and "legacy" not in model_name)
    ):
        return model_name + "-completion"
    else:
        return model_name


def get_openai_token_cost_for_model(
    model_name: str, num_tokens: int, is_completion: bool = False
) -> float:
    """
    Get the cost in USD for a given model and number of tokens.

    Args:
        model_name: Name of the model
        num_tokens: Number of tokens.
        is_completion: Whether the model is used for completion or not.
            Defaults to False.

    Returns:
        Cost in USD.
    """
    model_name = standardize_model_name(model_name, is_completion=is_completion)
    if model_name not in MODEL_COST_PER_1K_TOKENS:
        raise ValueError(
            f"Unknown model: {model_name}. Please provide a valid OpenAI model name."
            "Known models are: " + ", ".join(MODEL_COST_PER_1K_TOKENS.keys())
        )
    return MODEL_COST_PER_1K_TOKENS[model_name] * (num_tokens / 1000)


class OpenAICallbackHandler(BaseCallbackHandler):
    """Callback Handler that tracks OpenAI info."""

    total_tokens: int = 0
    prompt_tokens: int = 0
    completion_tokens: int = 0
    successful_requests: int = 0
    total_cost: float = 0.0

    def __init__(self) -> None:
        super().__init__()
        self._lock = threading.Lock()

    def __repr__(self) -> str:
        return (
            f"Tokens Used: {self.total_tokens}\n"
            f"\tPrompt Tokens: {self.prompt_tokens}\n"
            f"\tCompletion Tokens: {self.completion_tokens}\n"
            f"Successful Requests: {self.successful_requests}\n"
            f"Total Cost (USD): ${self.total_cost}"
        )

    @property
    def always_verbose(self) -> bool:
        """Whether to call verbose callbacks even if verbose is False."""
        return True

    def on_llm_start(
        self, serialized: Dict[str, Any], prompts: List[str], **kwargs: Any
    ) -> None:
        """Print out the prompts."""
        pass

    def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
        """Print out the token."""
        pass

    def on_llm_end(self, response: LLMResult, **kwargs: Any) -> None:
        """Collect token usage."""
        # Check for usage_metadata (langchain-core >= 0.2.2)
        try:
            generation = response.generations[0][0]
        except IndexError:
            generation = None
        if isinstance(generation, ChatGeneration):
            try:
                message = generation.message
                if isinstance(message, AIMessage):
                    usage_metadata = message.usage_metadata
                else:
                    usage_metadata = None
            except AttributeError:
                usage_metadata = None
        else:
            usage_metadata = None
        if usage_metadata:
            token_usage = {"total_tokens": usage_metadata["total_tokens"]}
            completion_tokens = usage_metadata["output_tokens"]
            prompt_tokens = usage_metadata["input_tokens"]
            if response.llm_output is None:
                # model name (and therefore cost) is unavailable in
                # streaming responses
                model_name = ""
            else:
                model_name = standardize_model_name(
                    response.llm_output.get("model_name", "")
                )

        else:
            if response.llm_output is None:
                return None

            if "token_usage" not in response.llm_output:
                with self._lock:
                    self.successful_requests += 1
                return None

            # compute tokens and cost for this request
            token_usage = response.llm_output["token_usage"]
            completion_tokens = token_usage.get("completion_tokens", 0)
            prompt_tokens = token_usage.get("prompt_tokens", 0)
            model_name = standardize_model_name(
                response.llm_output.get("model_name", "")
            )
        if model_name in MODEL_COST_PER_1K_TOKENS:
            completion_cost = get_openai_token_cost_for_model(
                model_name, completion_tokens, is_completion=True
            )
            prompt_cost = get_openai_token_cost_for_model(model_name, prompt_tokens)
        else:
            completion_cost = 0
            prompt_cost = 0

        # update shared state behind lock
        with self._lock:
            self.total_cost += prompt_cost + completion_cost
            self.total_tokens += token_usage.get("total_tokens", 0)
            self.prompt_tokens += prompt_tokens
            self.completion_tokens += completion_tokens
            self.successful_requests += 1

    def __copy__(self) -> "OpenAICallbackHandler":
        """Return a copy of the callback handler."""
        return self

    def __deepcopy__(self, memo: Any) -> "OpenAICallbackHandler":
        """Return a deep copy of the callback handler."""
        return self