langchain/libs/partners/fireworks/langchain_fireworks/llms.py

"""Wrapper around Fireworks AI's Completion API."""

import logging
from typing import Any, Dict, List, Optional

import requests
from aiohttp import ClientSession
from langchain_core.callbacks import (
    AsyncCallbackManagerForLLMRun,
    CallbackManagerForLLMRun,
)
from langchain_core.language_models.llms import LLM
from langchain_core.pydantic_v1 import Extra, Field, SecretStr, root_validator
from langchain_core.utils import (
    convert_to_secret_str,
    get_from_dict_or_env,
    get_pydantic_field_names,
)
from langchain_core.utils.utils import build_extra_kwargs

from langchain_fireworks.version import __version__

logger = logging.getLogger(__name__)


class Fireworks(LLM):
    """LLM models from `Fireworks`.

    To use, you'll need an API key which you can find here:
    https://fireworks.ai This can be passed in as init param
    ``fireworks_api_key`` or set as environment variable ``FIREWORKS_API_KEY``.

    Fireworks AI API reference: https://readme.fireworks.ai/

    Example:
        .. code-block:: python
            response = fireworks.generate(["Tell me a joke."])
    """

    base_url: str = "https://api.fireworks.ai/inference/v1/completions"
    """Base inference API URL."""
    fireworks_api_key: SecretStr = Field(default=None, alias="api_key")
    """Fireworks AI API key. Get it here: https://fireworks.ai"""
    model: str
    """Model name. Available models listed here: 
        https://readme.fireworks.ai/
    """
    temperature: Optional[float] = None
    """Model temperature."""
    top_p: Optional[float] = None
    """Used to dynamically adjust the number of choices for each predicted token based 
        on the cumulative probabilities. A value of 1 will always yield the same 
        output. A temperature less than 1 favors more correctness and is appropriate 
        for question answering or summarization. A value greater than 1 introduces more 
        randomness in the output.
    """
    model_kwargs: Dict[str, Any] = Field(default_factory=dict)
    """Holds any model parameters valid for `create` call not explicitly specified."""
    top_k: Optional[int] = None
    """Used to limit the number of choices for the next predicted word or token. It 
        specifies the maximum number of tokens to consider at each step, based on their 
        probability of occurrence. This technique helps to speed up the generation 
        process and can improve the quality of the generated text by focusing on the 
        most likely options.
    """
    max_tokens: Optional[int] = None
    """The maximum number of tokens to generate."""
    repetition_penalty: Optional[float] = None
    """A number that controls the diversity of generated text by reducing the 
        likelihood of repeated sequences. Higher values decrease repetition.
    """
    logprobs: Optional[int] = None
    """An integer that specifies how many top token log probabilities are included in 
        the response for each token generation step.
    """

    class Config:
        """Configuration for this pydantic object."""

        extra = Extra.forbid
        allow_population_by_field_name = True

    @root_validator(pre=True)
    def build_extra(cls, values: Dict[str, Any]) -> Dict[str, Any]:
        """Build extra kwargs from additional params that were passed in."""
        all_required_field_names = get_pydantic_field_names(cls)
        extra = values.get("model_kwargs", {})
        values["model_kwargs"] = build_extra_kwargs(
            extra, values, all_required_field_names
        )
        return values

    @root_validator(pre=True)
    def validate_environment(cls, values: Dict) -> Dict:
        """Validate that api key exists in environment."""
        values["fireworks_api_key"] = convert_to_secret_str(
            get_from_dict_or_env(values, "fireworks_api_key", "FIREWORKS_API_KEY")
        )
        return values

    @property
    def _llm_type(self) -> str:
        """Return type of model."""
        return "fireworks"

    def _format_output(self, output: dict) -> str:
        return output["choices"][0]["text"]

    @staticmethod
    def get_user_agent() -> str:
        return f"langchain-fireworks/{__version__}"

    @property
    def default_params(self) -> Dict[str, Any]:
        return {
            "model": self.model,
            "temperature": self.temperature,
            "top_p": self.top_p,
            "top_k": self.top_k,
            "max_tokens": self.max_tokens,
            "repetition_penalty": self.repetition_penalty,
        }

    def _call(
        self,
        prompt: str,
        stop: Optional[List[str]] = None,
        run_manager: Optional[CallbackManagerForLLMRun] = None,
        **kwargs: Any,
    ) -> str:
        """Call out to Fireworks's text generation endpoint.

        Args:
            prompt: The prompt to pass into the model.

        Returns:
            The string generated by the model..
        """
        headers = {
            "Authorization": f"Bearer {self.fireworks_api_key.get_secret_value()}",
            "Content-Type": "application/json",
        }
        stop_to_use = stop[0] if stop and len(stop) == 1 else stop
        payload: Dict[str, Any] = {
            **self.default_params,
            "prompt": prompt,
            "stop": stop_to_use,
            **kwargs,
        }

        # filter None values to not pass them to the http payload
        payload = {k: v for k, v in payload.items() if v is not None}
        response = requests.post(url=self.base_url, json=payload, headers=headers)

        if response.status_code >= 500:
            raise Exception(f"Fireworks Server: Error {response.status_code}")
        elif response.status_code >= 400:
            raise ValueError(f"Fireworks received an invalid payload: {response.text}")
        elif response.status_code != 200:
            raise Exception(
                f"Fireworks returned an unexpected response with status "
                f"{response.status_code}: {response.text}"
            )

        data = response.json()
        output = self._format_output(data)

        return output

    async def _acall(
        self,
        prompt: str,
        stop: Optional[List[str]] = None,
        run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
        **kwargs: Any,
    ) -> str:
        """Call Fireworks model to get predictions based on the prompt.

        Args:
            prompt: The prompt to pass into the model.

        Returns:
            The string generated by the model.
        """
        headers = {
            "Authorization": f"Bearer {self.fireworks_api_key.get_secret_value()}",
            "Content-Type": "application/json",
        }
        stop_to_use = stop[0] if stop and len(stop) == 1 else stop
        payload: Dict[str, Any] = {
            **self.default_params,
            "prompt": prompt,
            "stop": stop_to_use,
            **kwargs,
        }

        # filter None values to not pass them to the http payload
        payload = {k: v for k, v in payload.items() if v is not None}
        async with ClientSession() as session:
            async with session.post(
                self.base_url, json=payload, headers=headers
            ) as response:
                if response.status >= 500:
                    raise Exception(f"Fireworks Server: Error {response.status}")
                elif response.status >= 400:
                    raise ValueError(
                        f"Fireworks received an invalid payload: {response.text}"
                    )
                elif response.status != 200:
                    raise Exception(
                        f"Fireworks returned an unexpected response with status "
                        f"{response.status}: {response.text}"
                    )

                response_json = await response.json()
                output = self._format_output(response_json)
                return output