feat: Support gemini stt model

Author: liuruibin

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

@@ -0,0 +1,41 @@
+# coding=utf-8
+from typing import Dict
+
+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 GeminiSTTModelCredential(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)
+            model.check_auth()
+        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/gemini_model_provider/gemini_model_provider.py

@@ -13,12 +13,15 @@
     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.credential.stt import GeminiSTTModelCredential
 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 setting.models_provider.impl.gemini_model_provider.model.stt import GeminiSpeechToText
 from smartdoc.conf import PROJECT_DIR
 
 gemini_llm_model_credential = GeminiLLMModelCredential()
 gemini_image_model_credential = GeminiImageModelCredential()
+gemini_stt_model_credential = GeminiSTTModelCredential()
 
 model_info_list = [
     ModelInfo('gemini-1.0-pro', '最新的Gemini 1.0 Pro模型，随Google更新而更新',
@@ -42,14 +45,25 @@
               GeminiImage),
 ]
 
-
+model_stt_info_list = [
+    ModelInfo('gemini-1.5-flash', '最新的Gemini 1.5 Flash模型，随Google更新而更新',
+              ModelTypeConst.STT,
+              gemini_stt_model_credential,
+              GeminiSpeechToText),
+    ModelInfo('gemini-1.5-pro', '最新的Gemini 1.5 Flash模型，随Google更新而更新',
+              ModelTypeConst.STT,
+              gemini_stt_model_credential,
+              GeminiSpeechToText),
+]
 
 model_info_manage = (
     ModelInfoManage.builder()
     .append_model_info_list(model_info_list)
     .append_model_info_list(model_image_info_list)
+    .append_model_info_list(model_stt_info_list)
     .append_default_model_info(model_info_list[0])
     .append_default_model_info(model_image_info_list[0])
+    .append_default_model_info(model_stt_info_list[0])
     .build()
 )
 

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

@@ -0,0 +1,60 @@
+import asyncio
+import io
+from typing import Dict
+
+from langchain_core.messages import HumanMessage
+from langchain_google_genai import ChatGoogleGenerativeAI
+from openai import OpenAI
+
+from common.config.tokenizer_manage_config import TokenizerManage
+from setting.models_provider.base_model_provider import MaxKBBaseModel
+from setting.models_provider.impl.base_stt import BaseSpeechToText
+import google.generativeai as genai
+
+
+def custom_get_token_ids(text: str):
+    tokenizer = TokenizerManage.get_tokenizer()
+    return tokenizer.encode(text)
+
+
+class GeminiSpeechToText(MaxKBBaseModel, BaseSpeechToText):
+    api_key: str
+    model: str
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self.api_key = kwargs.get('api_key')
+
+    @staticmethod
+    def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs):
+        optional_params = {}
+        if 'max_tokens' in model_kwargs and model_kwargs['max_tokens'] is not None:
+            optional_params['max_tokens'] = model_kwargs['max_tokens']
+        if 'temperature' in model_kwargs and model_kwargs['temperature'] is not None:
+            optional_params['temperature'] = model_kwargs['temperature']
+        return GeminiSpeechToText(
+            model=model_name,
+            api_key=model_credential.get('api_key'),
+            **optional_params,
+        )
+
+    def check_auth(self):
+        client = ChatGoogleGenerativeAI(
+            model=self.model,
+            google_api_key=self.api_key
+        )
+        response_list = client.invoke('你好')
+        # print(response_list)
+
+    def speech_to_text(self, audio_file):
+        client = ChatGoogleGenerativeAI(
+            model=self.model,
+            google_api_key=self.api_key
+        )
+        audio_data = audio_file.read()
+        msg = HumanMessage(content=[
+            {'type': 'text', 'text': '把音频转成文字'},
+            {"type": "media", 'mime_type': 'audio/mp3', "data": audio_data}
+        ])
+        res = client.invoke([msg])
+        return res.content