feat: add baichuan llm support

api/core/model_providers/model_provider_factory.py

@@ -51,6 +51,9 @@ def get_model_provider_class(cls, provider_name: str) -> Type[BaseModelProvider]
         elif provider_name == 'chatglm':
             from core.model_providers.providers.chatglm_provider import ChatGLMProvider
             return ChatGLMProvider
+        elif provider_name == 'baichuan':
+            from core.model_providers.providers.baichuan_provider import BaichuanProvider
+            return BaichuanProvider
         elif provider_name == 'azure_openai':
             from core.model_providers.providers.azure_openai_provider import AzureOpenAIProvider
             return AzureOpenAIProvider

api/core/model_providers/models/llm/baichuan_model.py

@@ -0,0 +1,61 @@
+from typing import List, Optional, Any
+
+from langchain.callbacks.manager import Callbacks
+from langchain.schema import LLMResult
+
+from core.model_providers.error import LLMBadRequestError
+from core.model_providers.models.llm.base import BaseLLM
+from core.model_providers.models.entity.message import PromptMessage
+from core.model_providers.models.entity.model_params import ModelMode, ModelKwargs
+from core.third_party.langchain.llms.baichuan_llm import BaichuanChatLLM
+
+
+class BaichuanModel(BaseLLM):
+    model_mode: ModelMode = ModelMode.CHAT
+
+    def _init_client(self) -> Any:
+        provider_model_kwargs = self._to_model_kwargs_input(self.model_rules, self.model_kwargs)
+        return BaichuanChatLLM(
+            streaming=self.streaming,
+            callbacks=self.callbacks,
+            **self.credentials,
+            **provider_model_kwargs
+        )
+
+    def _run(self, messages: List[PromptMessage],
+             stop: Optional[List[str]] = None,
+             callbacks: Callbacks = None,
+             **kwargs) -> LLMResult:
+        """
+        run predict by prompt messages and stop words.
+
+        :param messages:
+        :param stop:
+        :param callbacks:
+        :return:
+        """
+        prompts = self._get_prompt_from_messages(messages)
+        return self._client.generate([prompts], stop, callbacks)
+
+    def get_num_tokens(self, messages: List[PromptMessage]) -> int:
+        """
+        get num tokens of prompt messages.
+
+        :param messages:
+        :return:
+        """
+        prompts = self._get_prompt_from_messages(messages)
+        return max(self._client.get_num_tokens_from_messages(prompts), 0)
+
+    def _set_model_kwargs(self, model_kwargs: ModelKwargs):
+        provider_model_kwargs = self._to_model_kwargs_input(self.model_rules, model_kwargs)
+        for k, v in provider_model_kwargs.items():
+            if hasattr(self.client, k):
+                setattr(self.client, k, v)
+
+    def handle_exceptions(self, ex: Exception) -> Exception:
+        return LLMBadRequestError(f"Baichuan: {str(ex)}")
+
+    @property
+    def support_streaming(self):
+        return True

api/core/model_providers/providers/baichuan_provider.py

@@ -0,0 +1,167 @@
+import json
+from json import JSONDecodeError
+from typing import Type
+
+from langchain.schema import HumanMessage
+
+from core.helper import encrypter
+from core.model_providers.models.base import BaseProviderModel
+from core.model_providers.models.entity.model_params import ModelKwargsRules, KwargRule, ModelType
+from core.model_providers.models.llm.baichuan_model import BaichuanModel
+from core.model_providers.providers.base import BaseModelProvider, CredentialsValidateFailedError
+from core.third_party.langchain.llms.baichuan_llm import BaichuanChatLLM
+from models.provider import ProviderType
+
+
+class BaichuanProvider(BaseModelProvider):
+
+    @property
+    def provider_name(self):
+        """
+        Returns the name of a provider.
+        """
+        return 'baichuan'
+
+    def _get_fixed_model_list(self, model_type: ModelType) -> list[dict]:
+        if model_type == ModelType.TEXT_GENERATION:
+            return [
+                {
+                    'id': 'baichuan2-53b',
+                    'name': 'Baichuan2-53B',
+                }
+            ]
+        else:
+            return []
+
+    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 = BaichuanModel
+        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:
+        """
+        return ModelKwargsRules(
+            temperature=KwargRule[float](min=0, max=1, default=0.3, precision=2),
+            top_p=KwargRule[float](min=0, max=0.99, default=0.85, precision=2),
+            presence_penalty=KwargRule[float](enabled=False),
+            frequency_penalty=KwargRule[float](enabled=False),
+            max_tokens=KwargRule[int](enabled=False),
+        )
+
+    @classmethod
+    def is_provider_credentials_valid_or_raise(cls, credentials: dict):
+        """
+        Validates the given credentials.
+        """
+        if 'api_key' not in credentials:
+            raise CredentialsValidateFailedError('Baichuan api_key must be provided.')
+
+        if 'secret_key' not in credentials:
+            raise CredentialsValidateFailedError('Baichuan secret_key must be provided.')
+
+        try:
+            credential_kwargs = {
+                'api_key': credentials['api_key'],
+                'secret_key': credentials['secret_key'],
+            }
+
+            llm = BaichuanChatLLM(
+                temperature=0,
+                **credential_kwargs
+            )
+
+            llm([HumanMessage(content='ping')])
+        except Exception as ex:
+            raise CredentialsValidateFailedError(str(ex))
+
+    @classmethod
+    def encrypt_provider_credentials(cls, tenant_id: str, credentials: dict) -> dict:
+        credentials['api_key'] = encrypter.encrypt_token(tenant_id, credentials['api_key'])
+        credentials['secret_key'] = encrypter.encrypt_token(tenant_id, credentials['secret_key'])
+        return credentials
+
+    def get_provider_credentials(self, obfuscated: bool = False) -> dict:
+        if self.provider.provider_type == ProviderType.CUSTOM.value:
+            try:
+                credentials = json.loads(self.provider.encrypted_config)
+            except JSONDecodeError:
+                credentials = {
+                    'api_key': None,
+                    'secret_key': None,
+                }
+
+            if credentials['api_key']:
+                credentials['api_key'] = encrypter.decrypt_token(
+                    self.provider.tenant_id,
+                    credentials['api_key']
+                )
+
+                if obfuscated:
+                    credentials['api_key'] = encrypter.obfuscated_token(credentials['api_key'])
+
+            if credentials['secret_key']:
+                credentials['secret_key'] = encrypter.decrypt_token(
+                    self.provider.tenant_id,
+                    credentials['secret_key']
+                )
+
+                if obfuscated:
+                    credentials['secret_key'] = encrypter.obfuscated_token(credentials['secret_key'])
+
+            return credentials
+        else:
+            return {}
+
+    def should_deduct_quota(self):
+        return True
+
+    @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:
+        """
+        return
+
+    @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:
+        """
+        return {}
+
+    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:
+        """
+        return self.get_provider_credentials(obfuscated)

api/core/model_providers/rules/_providers.json

@@ -7,10 +7,11 @@
   "spark",
   "wenxin",
   "zhipuai",
+  "baichuan",
   "chatglm",
   "replicate",
   "huggingface_hub",
   "xinference",
   "openllm",
   "localai"
-]
+]

api/core/model_providers/rules/baichuan.json

@@ -0,0 +1,15 @@
+{
+    "support_provider_types": [
+        "custom"
+    ],
+    "system_config": null,
+    "model_flexibility": "fixed",
+    "price_config": {
+        "baichuan2-53b": {
+            "prompt": "0.01",
+            "completion": "0.01",
+            "unit": "0.001",
+            "currency": "RMB"
+        }
+    }
+}