neollm/llm/gpt/azure_llm.py

from typing import Literal, cast

from openai import AzureOpenAI

from neollm.llm.abstract_llm import AbstractLLM
from neollm.llm.gpt.abstract_gpt import AbstractGPT
from neollm.types import (
    APIPricing,
    ClientSettings,
    LLMSettings,
    Messages,
    Response,
    StreamResponse,
)
from neollm.utils.utils import cprint, ensure_env_var, suport_unrecomended_env_var

suport_unrecomended_env_var(old_key="AZURE_API_BASE", new_key="AZURE_OPENAI_ENDPOINT")
suport_unrecomended_env_var(old_key="AZURE_API_VERSION", new_key="OPENAI_API_VERSION")
# 0613なし
suport_unrecomended_env_var(old_key="AZURE_ENGINE_GPT35", new_key="AZURE_ENGINE_GPT35T_0613")
suport_unrecomended_env_var(old_key="AZURE_ENGINE_GPT35_16k", new_key="AZURE_ENGINE_GPT35T_16K_0613")
suport_unrecomended_env_var(old_key="AZURE_ENGINE_GPT4", new_key="AZURE_ENGINE_GPT4_0613")
suport_unrecomended_env_var(old_key="AZURE_ENGINE_GPT4_32k", new_key="AZURE_ENGINE_GPT4_32K_0613")
# turbo抜け
suport_unrecomended_env_var(old_key="AZURE_ENGINE_GPT35_0613", new_key="AZURE_ENGINE_GPT35T_0613")
suport_unrecomended_env_var(old_key="AZURE_ENGINE_GPT35_16K_0613", new_key="AZURE_ENGINE_GPT35T_16K_0613")

# Pricing: https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/

SUPPORTED_MODELS = Literal[
    "gpt-4o-2024-05-13",
    "gpt-4-turbo-2024-04-09",
    "gpt-3.5-turbo-0125",
    "gpt-4-turbo-0125",
    "gpt-3.5-turbo-1106",
    "gpt-4-turbo-1106",
    "gpt-4v-turbo-1106",
    "gpt-3.5-turbo-0613",
    "gpt-3.5-turbo-16k-0613",
    "gpt-4-0613",
    "gpt-4-32k-0613",
]


def get_azure_llm(model_name: SUPPORTED_MODELS | str, client_settings: ClientSettings) -> AbstractLLM:
    # 表記変更
    model_name = model_name.replace("gpt-35-turbo", "gpt-3.5-turbo")
    # Add 日付
    replace_map_for_nodate: dict[str, SUPPORTED_MODELS] = {
        "gpt-4o": "gpt-4o-2024-05-13",
        "gpt-3.5-turbo": "gpt-3.5-turbo-0613",
        "gpt-3.5-turbo-16k": "gpt-3.5-turbo-16k-0613",
        "gpt-4": "gpt-4-0613",
        "gpt-4-32k": "gpt-4-32k-0613",
        "gpt-4-turbo": "gpt-4-turbo-1106",
        "gpt-4v-turbo": "gpt-4v-turbo-1106",
    }
    if model_name in replace_map_for_nodate:
        cprint("WARNING: model_nameに日付を指定してください", color="yellow", background=True)
        print(f"model_name: {model_name} -> {replace_map_for_nodate[model_name]}")
        model_name = replace_map_for_nodate[model_name]

    # map to LLM
    supported_model_map: dict[SUPPORTED_MODELS, AbstractLLM] = {
        "gpt-4o-2024-05-13": AzureGPT4O_20240513(client_settings),
        "gpt-4-turbo-2024-04-09": AzureGPT4T_20240409(client_settings),
        "gpt-3.5-turbo-0125": AzureGPT35T_0125(client_settings),
        "gpt-4-turbo-0125": AzureGPT4T_0125(client_settings),
        "gpt-3.5-turbo-1106": AzureGPT35T_1106(client_settings),
        "gpt-4-turbo-1106": AzureGPT4T_1106(client_settings),
        "gpt-4v-turbo-1106": AzureGPT4VT_1106(client_settings),
        "gpt-3.5-turbo-0613": AzureGPT35T_0613(client_settings),
        "gpt-3.5-turbo-16k-0613": AzureGPT35T16k_0613(client_settings),
        "gpt-4-0613": AzureGPT4_0613(client_settings),
        "gpt-4-32k-0613": AzureGPT432k_0613(client_settings),
    }
    # 通常モデル
    if model_name in supported_model_map:
        model_name = cast(SUPPORTED_MODELS, model_name)
        return supported_model_map[model_name]
    # FTモデル
    return AzureGPT35FT(model_name, client_settings)


class AzureLLM(AbstractGPT):
    _engine_name_env_key: str | None = None

    @property
    def client(self) -> AzureOpenAI:
        client: AzureOpenAI = AzureOpenAI(**self.client_settings)
        # api_key: str | None = (None,)
        # timeout: httpx.Timeout(timeout=600.0, connect=5.0)
        # max_retries: int = 2
        return client

    @property
    def engine(self) -> str:
        return ensure_env_var(self._engine_name_env_key)

    def generate(self, messages: Messages, llm_settings: LLMSettings) -> Response:
        openai_response = self.client.chat.completions.create(
            model=self.engine,
            messages=self._convert_to_platform_messages(messages),
            stream=False,
            **llm_settings,
        )
        response = self._convert_to_response(openai_response)
        return response

    def generate_stream(self, messages: Messages, llm_settings: LLMSettings) -> StreamResponse:
        platform_stream_response = self.client.chat.completions.create(
            model=self.engine,
            messages=self._convert_to_platform_messages(messages),
            stream=True,
            **llm_settings,
        )
        stream_response = self._convert_to_streamresponse(platform_stream_response)
        return stream_response


# omni 2024-05-13 --------------------------------------------------------------------------------------------
class AzureGPT4O_20240513(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.005, output=0.015)  # 30倍/45倍
    model: str = "gpt-4o-2024-05-13"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT4O_20240513"
    context_window: int = 128_000


# 2024-04-09 --------------------------------------------------------------------------------------------
class AzureGPT4T_20240409(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.01, output=0.03)
    model: str = "gpt-4-turbo-2024-04-09"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT4T_20240409"
    context_window: int = 128_000


# 0125 --------------------------------------------------------------------------------------------
class AzureGPT35T_0125(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.0005, output=0.0015)
    model: str = "gpt-3.5-turbo-0125"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT35T_0125"
    context_window: int = 16_385


class AzureGPT4T_0125(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.01, output=0.03)
    model: str = "gpt-4-turbo-0125"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT4T_0125"
    context_window: int = 128_000


# 1106 --------------------------------------------------------------------------------------------
class AzureGPT35T_1106(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.001, output=0.002)
    model: str = "gpt-3.5-turbo-1106"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT35T_1106"
    context_window: int = 16_385


class AzureGPT4VT_1106(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.01, output=0.03)  # 10倍/15倍
    model: str = "gpt-4-1106-vision-preview"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT4VT_1106"
    context_window: int = 128_000


class AzureGPT4T_1106(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.01, output=0.03)
    model: str = "gpt-4-turbo-1106"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT4T_1106"
    context_window: int = 128_000


# FT --------------------------------------------------------------------------------------------
class AzureGPT35FT(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.0005, output=0.0015)  # 1倍 + セッション稼働時間
    model: str = "gpt-3.5-turbo-ft"
    context_window: int = 4_096

    def __init__(self, model_name: str, client_setting: ClientSettings) -> None:
        super().__init__(client_setting)
        self._engine = model_name

    @property
    def engine(self) -> str:
        return self._engine


# 0613 --------------------------------------------------------------------------------------------
class AzureGPT35T_0613(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.0015, output=0.002)
    model: str = "gpt-3.5-turbo-0613"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT35T_0613"
    context_window: int = 4_096


class AzureGPT35T16k_0613(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.003, output=0.004)  # 2倍
    model: str = "gpt-3.5-turbo-16k-0613"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT35T_16K_0613"
    context_window: int = 16_385


class AzureGPT4_0613(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.03, output=0.06)  # 20倍/30倍
    model: str = "gpt-4-0613"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT4_0613"
    context_window: int = 8_192


class AzureGPT432k_0613(AzureLLM):
    dollar_per_ktoken = APIPricing(input=0.06, output=0.12)  # 40倍/60倍
    model: str = "gpt-4-32k-0613"
    _engine_name_env_key: str = "AZURE_ENGINE_GPT4_32K_0613"
    context_window: int = 32_768