feat: BaiLian Image Model

apps/setting/models_provider/impl/aliyun_bai_lian_model_provider/aliyun_bai_lian_model_provider.py

@@ -13,15 +13,19 @@ from setting.models_provider.base_model_provider import ModelProvideInfo, ModelT
     ModelInfoManage
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.credential.embedding import \
     AliyunBaiLianEmbeddingCredential
+from setting.models_provider.impl.aliyun_bai_lian_model_provider.credential.image import QwenVLModelCredential
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.credential.llm import BaiLianLLMModelCredential
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.credential.reranker import \
     AliyunBaiLianRerankerCredential
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.credential.stt import AliyunBaiLianSTTModelCredential
+from setting.models_provider.impl.aliyun_bai_lian_model_provider.credential.tti import QwenTextToImageModelCredential
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.credential.tts import AliyunBaiLianTTSModelCredential
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.model.embedding import AliyunBaiLianEmbedding
+from setting.models_provider.impl.aliyun_bai_lian_model_provider.model.image import QwenVLChatModel
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.model.llm import BaiLianChatModel
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.model.reranker import AliyunBaiLianReranker
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.model.stt import AliyunBaiLianSpeechToText
+from setting.models_provider.impl.aliyun_bai_lian_model_provider.model.tti import QwenTextToImageModel
 from setting.models_provider.impl.aliyun_bai_lian_model_provider.model.tts import AliyunBaiLianTextToSpeech
 from smartdoc.conf import PROJECT_DIR
 
@@ -30,6 +34,8 @@ aliyun_bai_lian_tts_model_credential = AliyunBaiLianTTSModelCredential()
 aliyun_bai_lian_stt_model_credential = AliyunBaiLianSTTModelCredential()
 aliyun_bai_lian_embedding_model_credential = AliyunBaiLianEmbeddingCredential()
 aliyun_bai_lian_llm_model_credential = BaiLianLLMModelCredential()
+qwenvl_model_credential = QwenVLModelCredential()
+qwentti_model_credential = QwenTextToImageModelCredential()
 
 model_info_list = [ModelInfo('gte-rerank',
                              '阿里巴巴通义实验室开发的GTE-Rerank文本排序系列模型，开发者可以通过LlamaIndex框架进行集成高质量文本检索、排序。',
@@ -52,9 +58,28 @@ model_info_list = [ModelInfo('gte-rerank',
                              BaiLianChatModel)
                    ]
 
-model_info_manage = ModelInfoManage.builder().append_model_info_list(model_info_list).append_default_model_info(
+module_info_vl_list = [
-    model_info_list[1]).append_default_model_info(model_info_list[2]).append_default_model_info(
+    ModelInfo('qwen-vl-max', '', ModelTypeConst.IMAGE, qwenvl_model_credential, QwenVLChatModel),
-    model_info_list[3]).append_default_model_info(model_info_list[4]).build()
+    ModelInfo('qwen-vl-max-0809', '', ModelTypeConst.IMAGE, qwenvl_model_credential, QwenVLChatModel),
+    ModelInfo('qwen-vl-plus-0809', '', ModelTypeConst.IMAGE, qwenvl_model_credential, QwenVLChatModel),
+]
+module_info_tti_list = [
+    ModelInfo('wanx-v1',
+              '通义万相-文本生成图像大模型，支持中英文双语输入，支持输入参考图片进行参考内容或者参考风格迁移，重点风格包括但不限于水彩、油画、中国画、素描、扁平插画、二次元、3D卡通。',
+              ModelTypeConst.TTI, qwentti_model_credential, QwenTextToImageModel),
+]
+
+model_info_manage = (
+    ModelInfoManage.builder()
+    .append_model_info_list(model_info_list)
+    .append_model_info_list(module_info_vl_list)
+    .append_model_info_list(module_info_tti_list)
+    .append_default_model_info(model_info_list[1])
+    .append_default_model_info(model_info_list[2])
+    .append_default_model_info(model_info_list[3])
+    .append_default_model_info(model_info_list[4])
+    .build()
+)
 
 
 class AliyunBaiLianModelProvider(IModelProvider):

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

@@ -0,0 +1,71 @@
+# coding=utf-8
+"""
+    @project: MaxKB
+    @Author：虎
+    @file： llm.py
+    @date：2024/7/11 18:41
+    @desc:
+"""
+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 QwenModelParams(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=1.0,
+                                    _min=0.1,
+                                    _max=1.9,
+                                    _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 QwenVLModelCredential(BaseForm, BaseModelCredential):
+
+    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', ''))}
+
+    api_key = forms.PasswordInputField('API Key', required=True)
+
+    def get_model_params_setting_form(self, model_name):
+        return QwenModelParams()

apps/setting/models_provider/impl/aliyun_bai_lian_model_provider/credential/tti.py

@@ -0,0 +1,94 @@
+# coding=utf-8
+"""
+    @project: MaxKB
+    @Author：虎
+    @file： llm.py
+    @date：2024/7/11 18:41
+    @desc:
+"""
+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 QwenModelParams(BaseForm):
+    size = forms.SingleSelect(
+        TooltipLabel('图片尺寸', '指定生成图片的尺寸, 如: 1024x1024'),
+        required=True,
+        default_value='1024*1024',
+        option_list=[
+            {'value': '1024*1024', 'label': '1024*1024'},
+            {'value': '720*1280', 'label': '720*1280'},
+            {'value': '768*1152', 'label': '768*1152'},
+            {'value': '1280*720', 'label': '1280*720'},
+        ],
+        text_field='label',
+        value_field='value')
+    n = forms.SliderField(
+        TooltipLabel('图片数量', '指定生成图片的数量'),
+        required=True, default_value=1,
+        _min=1,
+        _max=4,
+        _step=1,
+        precision=0)
+    style = forms.SingleSelect(
+        TooltipLabel('风格', '指定生成图片的风格'),
+        required=True,
+        default_value='<auto>',
+        option_list=[
+            {'value': '<auto>', 'label': '默认值，由模型随机输出图像风格'},
+            {'value': '<photography>', 'label': '摄影'},
+            {'value': '<portrait>', 'label': '人像写真'},
+            {'value': '<3d cartoon>', 'label': '3D卡通'},
+            {'value': '<anime>', 'label': '动画'},
+            {'value': '<oil painting>', 'label': '油画'},
+            {'value': '<watercolor>', 'label': '水彩'},
+            {'value': '<sketch>', 'label': '素描'},
+            {'value': '<chinese painting>', 'label': '中国画'},
+            {'value': '<flat illustration>', 'label': '扁平插画'},
+        ],
+        text_field='label',
+        value_field='value'
+    )
+
+
+class QwenTextToImageModelCredential(BaseForm, BaseModelCredential):
+
+    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.check_auth()
+            print(res)
+        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', ''))}
+
+    api_key = forms.PasswordInputField('API Key', required=True)
+
+    def get_model_params_setting_form(self, model_name):
+        return QwenModelParams()

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

@@ -0,0 +1,23 @@
+# coding=utf-8
+
+from typing import Dict
+
+from langchain_community.chat_models import ChatOpenAI
+
+from setting.models_provider.base_model_provider import MaxKBBaseModel
+
+
+class QwenVLChatModel(MaxKBBaseModel, ChatOpenAI):
+
+    @staticmethod
+    def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs):
+        optional_params = MaxKBBaseModel.filter_optional_params(model_kwargs)
+        chat_tong_yi = QwenVLChatModel(
+            model_name=model_name,
+            openai_api_key=model_credential.get('api_key'),
+            openai_api_base='https://dashscope.aliyuncs.com/compatible-mode/v1',
+            # stream_options={"include_usage": True},
+            streaming=True,
+            model_kwargs=optional_params,
+        )
+        return chat_tong_yi

apps/setting/models_provider/impl/aliyun_bai_lian_model_provider/model/tti.py

@@ -0,0 +1,58 @@
+# coding=utf-8
+from http import HTTPStatus
+from typing import Dict
+
+from dashscope import ImageSynthesis
+from langchain_community.chat_models import ChatTongyi
+from langchain_core.messages import HumanMessage
+
+from setting.models_provider.base_model_provider import MaxKBBaseModel
+from setting.models_provider.impl.base_tti import BaseTextToImage
+
+
+class QwenTextToImageModel(MaxKBBaseModel, BaseTextToImage):
+    api_key: str
+    model_name: str
+    params: dict
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self.api_key = kwargs.get('api_key')
+        self.model_name = kwargs.get('model_name')
+        self.params = kwargs.get('params')
+
+    @staticmethod
+    def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs):
+        optional_params = {'params': {'size': '1024*1024', 'style': '<auto>', 'n': 1}}
+        for key, value in model_kwargs.items():
+            if key not in ['model_id', 'use_local', 'streaming']:
+                optional_params['params'][key] = value
+        chat_tong_yi = QwenTextToImageModel(
+            model_name=model_name,
+            api_key=model_credential.get('api_key'),
+            **optional_params,
+        )
+        return chat_tong_yi
+
+    def is_cache_model(self):
+        return False
+
+    def check_auth(self):
+        chat = ChatTongyi(api_key=self.api_key, model_name='qwen-max')
+        chat.invoke([HumanMessage([{"type": "text", "text": "你好"}])])
+
+    def generate_image(self, prompt: str, negative_prompt: str = None):
+        # api_base='https://dashscope.aliyuncs.com/compatible-mode/v1',
+        rsp = ImageSynthesis.call(api_key=self.api_key,
+                                  model=self.model_name,
+                                  prompt=prompt,
+                                  negative_prompt=negative_prompt,
+                                  **self.params)
+        file_urls = []
+        if rsp.status_code == HTTPStatus.OK:
+            for result in rsp.output.results:
+                file_urls.append(result.url)
+        else:
+            print('sync_call Failed, status_code: %s, code: %s, message: %s' %
+                  (rsp.status_code, rsp.code, rsp.message))
+        return file_urls