dify/api/core/model_providers/providers/xinference_provider.py

import json
from typing import Type

import requests
from xinference_client.client.restful.restful_client import Client

from core.helper import encrypter
from core.model_providers.models.embedding.xinference_embedding import XinferenceEmbedding
from core.model_providers.models.entity.model_params import KwargRule, ModelKwargsRules, ModelType, ModelMode
from core.model_providers.models.llm.xinference_model import XinferenceModel
from core.model_providers.models.reranking.xinference_reranking import XinferenceReranking
from core.model_providers.providers.base import BaseModelProvider, CredentialsValidateFailedError

from core.model_providers.models.base import BaseProviderModel
from core.third_party.langchain.embeddings.xinference_embedding import XinferenceEmbeddings
from core.third_party.langchain.llms.xinference_llm import XinferenceLLM
from models.provider import ProviderType


class XinferenceProvider(BaseModelProvider):
    @property
    def provider_name(self):
        """
        Returns the name of a provider.
        """
        return 'xinference'

    def _get_fixed_model_list(self, model_type: ModelType) -> list[dict]:
        return []

    def _get_text_generation_model_mode(self, model_name) -> str:
        return ModelMode.COMPLETION.value

    def get_model_class(self, model_type: ModelType) -> Type[BaseProviderModel]:
        """
        Returns the model class.

        :param model_type:
        :return:
        """
        if model_type == ModelType.TEXT_GENERATION:
            model_class = XinferenceModel
        elif model_type == ModelType.EMBEDDINGS:
            model_class = XinferenceEmbedding
        elif model_type == ModelType.RERANKING:
            model_class = XinferenceReranking
        else:
            raise NotImplementedError

        return model_class

    def get_model_parameter_rules(self, model_name: str, model_type: ModelType) -> ModelKwargsRules:
        """
        get model parameter rules.

        :param model_name:
        :param model_type:
        :return:
        """
        credentials = self.get_model_credentials(model_name, model_type)
        if credentials['model_format'] == "ggmlv3" and credentials["model_handle_type"] == "chatglm":
            return ModelKwargsRules(
                temperature=KwargRule[float](min=0.01, max=2, default=1, precision=2),
                top_p=KwargRule[float](min=0, max=1, default=0.7, precision=2),
                presence_penalty=KwargRule[float](enabled=False),
                frequency_penalty=KwargRule[float](enabled=False),
                max_tokens=KwargRule[int](min=10, max=4000, default=256, precision=0),
            )
        elif credentials['model_format'] == "ggmlv3":
            return ModelKwargsRules(
                temperature=KwargRule[float](min=0.01, max=2, default=1, precision=2),
                top_p=KwargRule[float](min=0, max=1, default=0.7, precision=2),
                presence_penalty=KwargRule[float](min=-2, max=2, default=0, precision=2),
                frequency_penalty=KwargRule[float](min=-2, max=2, default=0, precision=2),
                max_tokens=KwargRule[int](min=10, max=4000, default=256, precision=0),
            )
        else:
            return ModelKwargsRules(
                temperature=KwargRule[float](min=0.01, max=2, default=1, precision=2),
                top_p=KwargRule[float](min=0, max=1, default=0.7, precision=2),
                presence_penalty=KwargRule[float](enabled=False),
                frequency_penalty=KwargRule[float](enabled=False),
                max_tokens=KwargRule[int](min=10, max=4000, default=256, precision=0),
            )


    @classmethod
    def is_model_credentials_valid_or_raise(cls, model_name: str, model_type: ModelType, credentials: dict):
        """
        check model credentials valid.

        :param model_name:
        :param model_type:
        :param credentials:
        """
        if 'server_url' not in credentials:
            raise CredentialsValidateFailedError('Xinference Server URL must be provided.')

        if 'model_uid' not in credentials:
            raise CredentialsValidateFailedError('Xinference Model UID must be provided.')

        try:
            credential_kwargs = {
                'server_url': credentials['server_url'],
                'model_uid': credentials['model_uid'],
            }

            if model_type == ModelType.TEXT_GENERATION:
                llm = XinferenceLLM(
                    **credential_kwargs
                )

                llm("ping")
            elif model_type == ModelType.EMBEDDINGS:
                embedding = XinferenceEmbeddings(
                    **credential_kwargs
                )

                embedding.embed_query("ping")
            elif model_type == ModelType.RERANKING:
                rerank_client = Client(credential_kwargs['server_url'])
                model = rerank_client.get_model(credential_kwargs['model_uid'])
                model.rerank(query="ping", documents=["ping", "pong"], top_n=2)
        except Exception as ex:
            raise CredentialsValidateFailedError(str(ex))

    @classmethod
    def encrypt_model_credentials(cls, tenant_id: str, model_name: str, model_type: ModelType,
                                  credentials: dict) -> dict:
        """
        encrypt model credentials for save.

        :param tenant_id:
        :param model_name:
        :param model_type:
        :param credentials:
        :return:
        """
        if model_type == ModelType.TEXT_GENERATION:
            extra_credentials = cls._get_extra_credentials(credentials)
            credentials.update(extra_credentials)

        credentials['server_url'] = encrypter.encrypt_token(tenant_id, credentials['server_url'])

        return credentials

    def get_model_credentials(self, model_name: str, model_type: ModelType, obfuscated: bool = False) -> dict:
        """
        get credentials for llm use.

        :param model_name:
        :param model_type:
        :param obfuscated:
        :return:
        """
        if self.provider.provider_type != ProviderType.CUSTOM.value:
            raise NotImplementedError

        provider_model = self._get_provider_model(model_name, model_type)

        if not provider_model.encrypted_config:
            return {
                'server_url': None,
                'model_uid': None,
            }

        credentials = json.loads(provider_model.encrypted_config)
        if credentials['server_url']:
            credentials['server_url'] = encrypter.decrypt_token(
                self.provider.tenant_id,
                credentials['server_url']
            )

            if obfuscated:
                credentials['server_url'] = encrypter.obfuscated_token(credentials['server_url'])

        return credentials

    @classmethod
    def _get_extra_credentials(self, credentials: dict) -> dict:
        url = f"{credentials['server_url']}/v1/models/{credentials['model_uid']}"
        response = requests.get(url)
        if response.status_code != 200:
            raise RuntimeError(
                f"Failed to get the model description, detail: {response.json()['detail']}"
            )
        desc = response.json()

        extra_credentials = {
            'model_format': desc['model_format'],
        }
        if desc["model_format"] == "ggmlv3" and "chatglm" in desc["model_name"]:
            extra_credentials['model_handle_type'] = 'chatglm'
        elif "generate" in desc["model_ability"]:
            extra_credentials['model_handle_type'] = 'generate'
        elif "chat" in desc["model_ability"]:
            extra_credentials['model_handle_type'] = 'chat'
        else:
            raise NotImplementedError(f"Model handle type not supported.")

        return extra_credentials

    @classmethod
    def is_provider_credentials_valid_or_raise(cls, credentials: dict):
        return

    @classmethod
    def encrypt_provider_credentials(cls, tenant_id: str, credentials: dict) -> dict:
        return {}

    def get_provider_credentials(self, obfuscated: bool = False) -> dict:
        return {}