feat: Gemini Image understand model

Author: liuruibin

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

@@ -0,0 +1,64 @@
+# 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, TooltipLabel
+from setting.models_provider.base_model_provider import BaseModelCredential, ValidCode
+
+class GeminiImageModelParams(BaseForm):
+    temperature = forms.SliderField(TooltipLabel('温度', '较高的数值会使输出更加随机，而较低的数值会使其更加集中和确定'),
+                                    required=True, default_value=0.7,
+                                    _min=0.1,
+                                    _max=1.0,
+                                    _step=0.01,
+                                    precision=2)
+
+    max_tokens = forms.SliderField(
+        TooltipLabel('输出最大Tokens', '指定模型可生成的最大token个数'),
+        required=True, default_value=800,
+        _min=1,
+        _max=100000,
+        _step=1,
+        precision=0)
+
+
+
+class GeminiImageModelCredential(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):
+        return GeminiImageModelParams()

apps/setting/models_provider/impl/gemini_model_provider/gemini_model_provider.py

@@ -11,24 +11,47 @@
 from common.util.file_util import get_file_content
 from setting.models_provider.base_model_provider import IModelProvider, ModelProvideInfo, ModelInfo, ModelTypeConst, \
     ModelInfoManage
+from setting.models_provider.impl.gemini_model_provider.credential.image import GeminiImageModelCredential
 from setting.models_provider.impl.gemini_model_provider.credential.llm import GeminiLLMModelCredential
+from setting.models_provider.impl.gemini_model_provider.model.image import GeminiImage
 from setting.models_provider.impl.gemini_model_provider.model.llm import GeminiChatModel
 from smartdoc.conf import PROJECT_DIR
 
 gemini_llm_model_credential = GeminiLLMModelCredential()
+gemini_image_model_credential = GeminiImageModelCredential()
 
-gemini_1_pro = ModelInfo('gemini-1.0-pro', '最新的Gemini 1.0 Pro模型，随Google更新而更新',
-                         ModelTypeConst.LLM,
-                         gemini_llm_model_credential,
-                         GeminiChatModel)
+model_info_list = [
+    ModelInfo('gemini-1.0-pro', '最新的Gemini 1.0 Pro模型，随Google更新而更新',
+              ModelTypeConst.LLM,
+              gemini_llm_model_credential,
+              GeminiChatModel),
+    ModelInfo('gemini-1.0-pro-vision', '最新的Gemini 1.0 Pro Vision模型，随Google更新而更新',
+              ModelTypeConst.LLM,
+              gemini_llm_model_credential,
+              GeminiChatModel),
+]
 
-gemini_1_pro_vision = ModelInfo('gemini-1.0-pro-vision', '最新的Gemini 1.0 Pro Vision模型，随Google更新而更新',
-                                ModelTypeConst.LLM,
-                                gemini_llm_model_credential,
-                                GeminiChatModel)
+model_image_info_list = [
+    ModelInfo('gemini-1.5-flash', '最新的Gemini 1.5 Flash模型，随Google更新而更新',
+              ModelTypeConst.IMAGE,
+              gemini_image_model_credential,
+              GeminiImage),
+    ModelInfo('gemini-1.5-pro', '最新的Gemini 1.5 Flash模型，随Google更新而更新',
+              ModelTypeConst.IMAGE,
+              gemini_image_model_credential,
+              GeminiImage),
+]
 
-model_info_manage = ModelInfoManage.builder().append_model_info(gemini_1_pro).append_model_info(
-    gemini_1_pro_vision).append_default_model_info(gemini_1_pro).build()
+
+
+model_info_manage = (
+    ModelInfoManage.builder()
+    .append_model_info_list(model_info_list)
+    .append_model_info_list(model_image_info_list)
+    .append_default_model_info(model_info_list[0])
+    .append_default_model_info(model_image_info_list[0])
+    .build()
+)
 
 
 class GeminiModelProvider(IModelProvider):

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

@@ -0,0 +1,24 @@
+from typing import Dict
+
+from langchain_google_genai import ChatGoogleGenerativeAI
+
+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 GeminiImage(MaxKBBaseModel, ChatGoogleGenerativeAI):
+
+    @staticmethod
+    def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs):
+        optional_params = MaxKBBaseModel.filter_optional_params(model_kwargs)
+        return GeminiImage(
+            model=model_name,
+            google_api_key=model_credential.get('api_key'),
+            streaming=True,
+            **optional_params,
+        )