feat: 支持智谱图片理解模型

Author: liuruibin

apps/setting/models_provider/impl/zhipu_model_provider/credential/image.py

@@ -0,0 +1,47 @@
+# coding=utf-8
+import base64
+import os
+from typing import Dict
+
+from langchain_core.messages import HumanMessage
+
+from common import forms
+from common.exception.app_exception import AppApiException
+from common.forms import BaseForm
+from setting.models_provider.base_model_provider import BaseModelCredential, ValidCode
+
+
+class ZhiPuImageModelCredential(BaseForm, BaseModelCredential):
+    api_key = forms.PasswordInputField('API Key', required=True)
+
+    def is_valid(self, model_type: str, model_name, model_credential: Dict[str, object], provider,
+                 raise_exception=False):
+        model_type_list = provider.get_model_type_list()
+        if not any(list(filter(lambda mt: mt.get('value') == model_type, model_type_list))):
+            raise AppApiException(ValidCode.valid_error.value, f'{model_type} 模型类型不支持')
+
+        for key in ['api_key']:
+            if key not in model_credential:
+                if raise_exception:
+                    raise AppApiException(ValidCode.valid_error.value, f'{key} 字段为必填字段')
+                else:
+                    return False
+        try:
+            model = provider.get_model(model_type, model_name, model_credential)
+            res = model.stream([HumanMessage(content=[{"type": "text", "text": "你好"}])])
+            for chunk in res:
+                print(chunk)
+        except Exception as e:
+            if isinstance(e, AppApiException):
+                raise e
+            if raise_exception:
+                raise AppApiException(ValidCode.valid_error.value, f'校验失败,请检查参数是否正确: {str(e)}')
+            else:
+                return False
+        return True
+
+    def encryption_dict(self, model: Dict[str, object]):
+        return {**model, 'api_key': super().encryption(model.get('api_key', ''))}
+
+    def get_model_params_setting_form(self, model_name):
+        pass

apps/setting/models_provider/impl/zhipu_model_provider/model/image.py

@@ -0,0 +1,26 @@
+from typing import Dict
+
+from langchain_community.chat_models import ChatZhipuAI
+from langchain_openai.chat_models import ChatOpenAI
+
+from common.config.tokenizer_manage_config import TokenizerManage
+from setting.models_provider.base_model_provider import MaxKBBaseModel
+
+
+def custom_get_token_ids(text: str):
+    tokenizer = TokenizerManage.get_tokenizer()
+    return tokenizer.encode(text)
+
+
+class ZhiPuImage(MaxKBBaseModel, ChatZhipuAI):
+
+    @staticmethod
+    def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs):
+        optional_params = MaxKBBaseModel.filter_optional_params(model_kwargs)
+        return ZhiPuImage(
+            model_name=model_name,
+            api_key=model_credential.get('api_key'),
+            # stream_options={"include_usage": True},
+            streaming=True,
+            **optional_params,
+        )

apps/setting/models_provider/impl/zhipu_model_provider/zhipu_model_provider.py

@@ -11,18 +11,40 @@
 from common.util.file_util import get_file_content
 from setting.models_provider.base_model_provider import ModelProvideInfo, ModelTypeConst, ModelInfo, IModelProvider, \
     ModelInfoManage
+from setting.models_provider.impl.zhipu_model_provider.credential.image import ZhiPuImageModelCredential
 from setting.models_provider.impl.zhipu_model_provider.credential.llm import ZhiPuLLMModelCredential
+from setting.models_provider.impl.zhipu_model_provider.model.image import ZhiPuImage
 from setting.models_provider.impl.zhipu_model_provider.model.llm import ZhipuChatModel
 from smartdoc.conf import PROJECT_DIR
 
 qwen_model_credential = ZhiPuLLMModelCredential()
+zhipu_image_model_credential = ZhiPuImageModelCredential()
+
 model_info_list = [
     ModelInfo('glm-4', '', ModelTypeConst.LLM, qwen_model_credential, ZhipuChatModel),
     ModelInfo('glm-4v', '', ModelTypeConst.LLM, qwen_model_credential, ZhipuChatModel),
     ModelInfo('glm-3-turbo', '', ModelTypeConst.LLM, qwen_model_credential, ZhipuChatModel)
 ]
-model_info_manage = ModelInfoManage.builder().append_model_info_list(model_info_list).append_default_model_info(
-    ModelInfo('glm-4', '', ModelTypeConst.LLM, qwen_model_credential, ZhipuChatModel)).build()
+
+model_info_image_list = [
+    ModelInfo('glm-4v-plus', '具有强大的多模态理解能力。能够同时理解多达五张图像，并支持视频内容理解',
+              ModelTypeConst.IMAGE, zhipu_image_model_credential,
+              ZhiPuImage),
+    ModelInfo('glm-4v', '专注于单图理解。适用于需要高效图像解析的场景',
+              ModelTypeConst.IMAGE, zhipu_image_model_credential,
+              ZhiPuImage),
+    ModelInfo('glm-4v-flash', '专注于单图理解。适用于需要高效图像解析的场景(免费)',
+              ModelTypeConst.IMAGE, zhipu_image_model_credential,
+              ZhiPuImage),
+]
+
+model_info_manage = (
+    ModelInfoManage.builder()
+    .append_model_info_list(model_info_list)
+    .append_default_model_info(ModelInfo('glm-4', '', ModelTypeConst.LLM, qwen_model_credential, ZhipuChatModel))
+    .append_model_info_list(model_info_image_list)
+    .build()
+)
 
 
 class ZhiPuModelProvider(IModelProvider):