feat: add zhipuai (#1188)

api/core/model_providers/model_provider_factory.py

@@ -45,6 +45,9 @@ class ModelProviderFactory:
         elif provider_name == 'wenxin':
             from core.model_providers.providers.wenxin_provider import WenxinProvider
             return WenxinProvider
+        elif provider_name == 'zhipuai':
+            from core.model_providers.providers.zhipuai_provider import ZhipuAIProvider
+            return ZhipuAIProvider
         elif provider_name == 'chatglm':
             from core.model_providers.providers.chatglm_provider import ChatGLMProvider
             return ChatGLMProvider

api/core/model_providers/models/embedding/zhipuai_embedding.py

@@ -0,0 +1,22 @@
+from core.model_providers.error import LLMBadRequestError
+from core.model_providers.providers.base import BaseModelProvider
+from core.model_providers.models.embedding.base import BaseEmbedding
+from core.third_party.langchain.embeddings.zhipuai_embedding import ZhipuAIEmbeddings
+
+
+class ZhipuAIEmbedding(BaseEmbedding):
+    def __init__(self, model_provider: BaseModelProvider, name: str):
+        credentials = model_provider.get_model_credentials(
+            model_name=name,
+            model_type=self.type
+        )
+
+        client = ZhipuAIEmbeddings(
+            model=name,
+            **credentials,
+        )
+
+        super().__init__(model_provider, client, name)
+
+    def handle_exceptions(self, ex: Exception) -> Exception:
+        return LLMBadRequestError(f"ZhipuAI embedding: {str(ex)}")

api/core/model_providers/models/entity/model_params.py

@@ -49,6 +49,7 @@ class KwargRule(Generic[T], BaseModel):
     max: Optional[T] = None
     default: Optional[T] = None
     alias: Optional[str] = None
+    precision: Optional[int] = None
 
 
 class ModelKwargsRules(BaseModel):

api/core/model_providers/models/llm/zhipuai_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.zhipuai_llm import ZhipuAIChatLLM
+
+
+class ZhipuAIModel(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 ZhipuAIChatLLM(
+            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"ZhipuAI: {str(ex)}")
+
+    @property
+    def support_streaming(self):
+        return True

api/core/model_providers/models/moderation/openai_moderation.py

@@ -23,14 +23,18 @@ class OpenAIModeration(BaseModeration):
 
         # 2000 text per chunk
         length = 2000
-        chunks = [text[i:i + length] for i in range(0, len(text), length)]
+        text_chunks = [text[i:i + length] for i in range(0, len(text), length)]
 
-        moderation_result = self._client.create(input=chunks,
-                                                api_key=credentials['openai_api_key'])
+        max_text_chunks = 32
+        chunks = [text_chunks[i:i + max_text_chunks] for i in range(0, len(text_chunks), max_text_chunks)]
 
-        for result in moderation_result.results:
-            if result['flagged'] is True:
-                return False
+        for text_chunk in chunks:
+            moderation_result = self._client.create(input=text_chunk,
+                                                    api_key=credentials['openai_api_key'])
+
+            for result in moderation_result.results:
+                if result['flagged'] is True:
+                    return False
 
         return True
 

api/core/model_providers/providers/anthropic_provider.py

@@ -69,11 +69,11 @@ class AnthropicProvider(BaseModelProvider):
         :return:
         """
         return ModelKwargsRules(
-            temperature=KwargRule[float](min=0, max=1, default=1),
-            top_p=KwargRule[float](min=0, max=1, default=0.7),
+            temperature=KwargRule[float](min=0, max=1, 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](alias="max_tokens_to_sample", min=10, max=100000, default=256),
+            max_tokens=KwargRule[int](alias="max_tokens_to_sample", min=10, max=100000, default=256, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/azure_openai_provider.py

@@ -164,14 +164,14 @@ class AzureOpenAIProvider(BaseModelProvider):
         model_credentials = self.get_model_credentials(model_name, model_type)
 
         return ModelKwargsRules(
-            temperature=KwargRule[float](min=0, max=2, default=1),
-            top_p=KwargRule[float](min=0, max=1, default=1),
-            presence_penalty=KwargRule[float](min=-2, max=2, default=0),
-            frequency_penalty=KwargRule[float](min=-2, max=2, default=0),
+            temperature=KwargRule[float](min=0, max=2, default=1, precision=2),
+            top_p=KwargRule[float](min=0, max=1, default=1, 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=base_model_max_tokens.get(
                 model_credentials['base_model_name'],
                 4097
-            ), default=16),
+            ), default=16, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/chatglm_provider.py

@@ -64,11 +64,11 @@ class ChatGLMProvider(BaseModelProvider):
         }
 
         return ModelKwargsRules(
-            temperature=KwargRule[float](min=0, max=2, default=1),
-            top_p=KwargRule[float](min=0, max=1, default=0.7),
+            temperature=KwargRule[float](min=0, 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](alias='max_token', min=10, max=model_max_tokens.get(model_name), default=2048),
+            max_tokens=KwargRule[int](alias='max_token', min=10, max=model_max_tokens.get(model_name), default=2048, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/hosted.py

@@ -45,6 +45,18 @@ class HostedModelProviders(BaseModel):
 hosted_model_providers = HostedModelProviders()
 
 
+class HostedModerationConfig(BaseModel):
+    enabled: bool = False
+    providers: list[str] = []
+
+
+class HostedConfig(BaseModel):
+    moderation = HostedModerationConfig()
+
+
+hosted_config = HostedConfig()
+
+
 def init_app(app: Flask):
     if os.environ.get("DEBUG") and os.environ.get("DEBUG").lower() == 'true':
         langchain.verbose = True
@@ -78,3 +90,9 @@ def init_app(app: Flask):
             paid_min_quantity=app.config.get("HOSTED_ANTHROPIC_PAID_MIN_QUANTITY"),
             paid_max_quantity=app.config.get("HOSTED_ANTHROPIC_PAID_MAX_QUANTITY"),
         )
+
+    if app.config.get("HOSTED_MODERATION_ENABLED") and app.config.get("HOSTED_MODERATION_PROVIDERS"):
+        hosted_config.moderation = HostedModerationConfig(
+            enabled=app.config.get("HOSTED_MODERATION_ENABLED"),
+            providers=app.config.get("HOSTED_MODERATION_PROVIDERS").split(',')
+        )

api/core/model_providers/providers/huggingface_hub_provider.py

@@ -47,11 +47,11 @@ class HuggingfaceHubProvider(BaseModelProvider):
         :return:
         """
         return ModelKwargsRules(
-            temperature=KwargRule[float](min=0, max=2, default=1),
-            top_p=KwargRule[float](min=0.01, max=0.99, default=0.7),
+            temperature=KwargRule[float](min=0, max=2, default=1, precision=2),
+            top_p=KwargRule[float](min=0.01, max=0.99, default=0.7, precision=2),
             presence_penalty=KwargRule[float](enabled=False),
             frequency_penalty=KwargRule[float](enabled=False),
-            max_tokens=KwargRule[int](alias='max_new_tokens', min=10, max=4000, default=200),
+            max_tokens=KwargRule[int](alias='max_new_tokens', min=10, max=4000, default=200, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/localai_provider.py

@@ -52,9 +52,9 @@ class LocalAIProvider(BaseModelProvider):
         :return:
         """
         return ModelKwargsRules(
-            temperature=KwargRule[float](min=0, max=2, default=0.7),
-            top_p=KwargRule[float](min=0, max=1, default=1),
-            max_tokens=KwargRule[int](min=10, max=4097, default=16),
+            temperature=KwargRule[float](min=0, max=2, default=0.7, precision=2),
+            top_p=KwargRule[float](min=0, max=1, default=1, precision=2),
+            max_tokens=KwargRule[int](min=10, max=4097, default=16, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/minimax_provider.py

@@ -74,11 +74,11 @@ class MinimaxProvider(BaseModelProvider):
         }
 
         return ModelKwargsRules(
-            temperature=KwargRule[float](min=0.01, max=1, default=0.9),
-            top_p=KwargRule[float](min=0, max=1, default=0.95),
+            temperature=KwargRule[float](min=0.01, max=1, default=0.9, precision=2),
+            top_p=KwargRule[float](min=0, max=1, default=0.95, precision=2),
             presence_penalty=KwargRule[float](enabled=False),
             frequency_penalty=KwargRule[float](enabled=False),
-            max_tokens=KwargRule[int](min=10, max=model_max_tokens.get(model_name, 6144), default=1024),
+            max_tokens=KwargRule[int](min=10, max=model_max_tokens.get(model_name, 6144), default=1024, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/openai_provider.py

@@ -133,11 +133,11 @@ class OpenAIProvider(BaseModelProvider):
         }
 
         return ModelKwargsRules(
-            temperature=KwargRule[float](min=0, max=2, default=1),
-            top_p=KwargRule[float](min=0, max=1, default=1),
-            presence_penalty=KwargRule[float](min=-2, max=2, default=0),
-            frequency_penalty=KwargRule[float](min=-2, max=2, default=0),
-            max_tokens=KwargRule[int](min=10, max=model_max_tokens.get(model_name, 4097), default=16),
+            temperature=KwargRule[float](min=0, max=2, default=1, precision=2),
+            top_p=KwargRule[float](min=0, max=1, default=1, 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=model_max_tokens.get(model_name, 4097), default=16, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/openllm_provider.py

@@ -45,11 +45,11 @@ class OpenLLMProvider(BaseModelProvider):
         :return:
         """
         return ModelKwargsRules(
-            temperature=KwargRule[float](min=0.01, max=2, default=1),
-            top_p=KwargRule[float](min=0, max=1, default=0.7),
-            presence_penalty=KwargRule[float](min=-2, max=2, default=0),
-            frequency_penalty=KwargRule[float](min=-2, max=2, default=0),
-            max_tokens=KwargRule[int](alias='max_new_tokens', min=10, max=4000, default=128),
+            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](alias='max_new_tokens', min=10, max=4000, default=128, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/replicate_provider.py

@@ -72,6 +72,7 @@ class ReplicateProvider(BaseModelProvider):
                         min=float(value.get('minimum')) if value.get('minimum') is not None else None,
                         max=float(value.get('maximum')) if value.get('maximum') is not None else None,
                         default=float(value.get('default')) if value.get('default') is not None else None,
+                        precision = 2
                     )
                     if key == 'temperature':
                         model_kwargs_rules.temperature = kwarg_rule
@@ -84,6 +85,7 @@ class ReplicateProvider(BaseModelProvider):
                         min=int(value.get('minimum')) if value.get('minimum') is not None else 1,
                         max=int(value.get('maximum')) if value.get('maximum') is not None else 8000,
                         default=int(value.get('default')) if value.get('default') is not None else 500,
+                        precision = 0
                     )
 
         return model_kwargs_rules

api/core/model_providers/providers/spark_provider.py

@@ -62,11 +62,11 @@ class SparkProvider(BaseModelProvider):
         :return:
         """
         return ModelKwargsRules(
-            temperature=KwargRule[float](min=0, max=1, default=0.5),
+            temperature=KwargRule[float](min=0, max=1, default=0.5, precision=2),
             top_p=KwargRule[float](enabled=False),
             presence_penalty=KwargRule[float](enabled=False),
             frequency_penalty=KwargRule[float](enabled=False),
-            max_tokens=KwargRule[int](min=10, max=4096, default=2048),
+            max_tokens=KwargRule[int](min=10, max=4096, default=2048, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/tongyi_provider.py

@@ -64,10 +64,10 @@ class TongyiProvider(BaseModelProvider):
 
         return ModelKwargsRules(
             temperature=KwargRule[float](enabled=False),
-            top_p=KwargRule[float](min=0, max=1, default=0.8),
+            top_p=KwargRule[float](min=0, max=1, default=0.8, precision=2),
             presence_penalty=KwargRule[float](enabled=False),
             frequency_penalty=KwargRule[float](enabled=False),
-            max_tokens=KwargRule[int](min=10, max=model_max_tokens.get(model_name), default=1024),
+            max_tokens=KwargRule[int](min=10, max=model_max_tokens.get(model_name), default=1024, precision=0),
         )
 
     @classmethod

api/core/model_providers/providers/wenxin_provider.py

@@ -63,8 +63,8 @@ class WenxinProvider(BaseModelProvider):
         """
         if model_name in ['ernie-bot', 'ernie-bot-turbo']:
             return ModelKwargsRules(
-                temperature=KwargRule[float](min=0.01, max=1, default=0.95),
-                top_p=KwargRule[float](min=0.01, max=1, default=0.8),
+                temperature=KwargRule[float](min=0.01, max=1, default=0.95, precision=2),
+                top_p=KwargRule[float](min=0.01, max=1, default=0.8, precision=2),
                 presence_penalty=KwargRule[float](enabled=False),
                 frequency_penalty=KwargRule[float](enabled=False),
                 max_tokens=KwargRule[int](enabled=False),

api/core/model_providers/providers/xinference_provider.py

@@ -53,27 +53,27 @@ class XinferenceProvider(BaseModelProvider):
         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),
-                top_p=KwargRule[float](min=0, max=1, default=0.7),
+                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),
+                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),
-                top_p=KwargRule[float](min=0, max=1, default=0.7),
-                presence_penalty=KwargRule[float](min=-2, max=2, default=0),
-                frequency_penalty=KwargRule[float](min=-2, max=2, default=0),
-                max_tokens=KwargRule[int](min=10, max=4000, default=256),
+                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),
-                top_p=KwargRule[float](min=0, max=1, default=0.7),
+                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),
+                max_tokens=KwargRule[int](min=10, max=4000, default=256, precision=0),
             )
 
 

api/core/model_providers/providers/zhipuai_provider.py

@@ -0,0 +1,176 @@
+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.embedding.zhipuai_embedding import ZhipuAIEmbedding
+from core.model_providers.models.entity.model_params import ModelKwargsRules, KwargRule, ModelType
+from core.model_providers.models.llm.zhipuai_model import ZhipuAIModel
+from core.model_providers.providers.base import BaseModelProvider, CredentialsValidateFailedError
+from core.third_party.langchain.llms.zhipuai_llm import ZhipuAIChatLLM
+from models.provider import ProviderType, ProviderQuotaType
+
+
+class ZhipuAIProvider(BaseModelProvider):
+
+    @property
+    def provider_name(self):
+        """
+        Returns the name of a provider.
+        """
+        return 'zhipuai'
+
+    def _get_fixed_model_list(self, model_type: ModelType) -> list[dict]:
+        if model_type == ModelType.TEXT_GENERATION:
+            return [
+                {
+                    'id': 'chatglm_pro',
+                    'name': 'chatglm_pro',
+                },
+                {
+                    'id': 'chatglm_std',
+                    'name': 'chatglm_std',
+                },
+                {
+                    'id': 'chatglm_lite',
+                    'name': 'chatglm_lite',
+                },
+                {
+                    'id': 'chatglm_lite_32k',
+                    'name': 'chatglm_lite_32k',
+                }
+            ]
+        elif model_type == ModelType.EMBEDDINGS:
+            return [
+                {
+                    'id': 'text_embedding',
+                    'name': 'text_embedding',
+                }
+            ]
+        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 = ZhipuAIModel
+        elif model_type == ModelType.EMBEDDINGS:
+            model_class = ZhipuAIEmbedding
+        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=0.01, max=1, default=0.95, precision=2),
+            top_p=KwargRule[float](min=0.1, max=0.9, default=0.8, precision=1),
+            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('ZhipuAI api_key must be provided.')
+
+        try:
+            credential_kwargs = {
+                'api_key': credentials['api_key']
+            }
+
+            llm = ZhipuAIChatLLM(
+                temperature=0.01,
+                **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'])
+        return credentials
+
+    def get_provider_credentials(self, obfuscated: bool = False) -> dict:
+        if self.provider.provider_type == ProviderType.CUSTOM.value \
+                or (self.provider.provider_type == ProviderType.SYSTEM.value
+                    and self.provider.quota_type == ProviderQuotaType.FREE.value):
+            try:
+                credentials = json.loads(self.provider.encrypted_config)
+            except JSONDecodeError:
+                credentials = {
+                    'api_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'])
+
+            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

@@ -6,6 +6,7 @@
   "tongyi",
   "spark",
   "wenxin",
+  "zhipuai",
   "chatglm",
   "replicate",
   "huggingface_hub",

api/core/model_providers/rules/zhipuai.json

@@ -0,0 +1,44 @@
+{
+    "support_provider_types": [
+        "system",
+        "custom"
+    ],
+    "system_config": {
+        "supported_quota_types": [
+            "free"
+        ],
+        "quota_unit": "tokens"
+    },
+    "model_flexibility": "fixed",
+    "price_config": {
+        "chatglm_pro": {
+            "prompt": "0.01",
+            "completion": "0.01",
+            "unit": "0.001",
+            "currency": "RMB"
+        },
+        "chatglm_std": {
+            "prompt": "0.005",
+            "completion": "0.005",
+            "unit": "0.001",
+            "currency": "RMB"
+        },
+        "chatglm_lite": {
+            "prompt": "0.002",
+            "completion": "0.002",
+            "unit": "0.001",
+            "currency": "RMB"
+        },
+        "chatglm_lite_32k": {
+            "prompt": "0.0004",
+            "completion": "0.0004",
+            "unit": "0.001",
+            "currency": "RMB"
+        },
+        "text_embedding": {
+            "completion": "0",
+            "unit": "0.001",
+            "currency": "RMB"
+        }
+    }
+}