feat: Xinference Image Model

Author: liuruibin

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

@@ -0,0 +1,65 @@
+# 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 XinferenceImageModelParams(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 XinferenceImageModelCredential(BaseForm, BaseModelCredential):
+    api_base = forms.TextInputField('API 域名', required=True)
+    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_base', '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 XinferenceImageModelParams()

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

@@ -0,0 +1,82 @@
+# 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 XinferenceTTIModelParams(BaseForm):
+    size = forms.SingleSelect(
+        TooltipLabel('图片尺寸', '指定生成图片的尺寸, 如: 1024x1024'),
+        required=True,
+        default_value='1024x1024',
+        option_list=[
+            {'value': '1024x1024', 'label': '1024x1024'},
+            {'value': '1024x1792', 'label': '1024x1792'},
+            {'value': '1792x1024', 'label': '1792x1024'},
+        ],
+        text_field='label',
+        value_field='value'
+    )
+
+    quality = forms.SingleSelect(
+        TooltipLabel('图片质量', ''),
+        required=True,
+        default_value='standard',
+        option_list=[
+            {'value': 'standard', 'label': 'standard'},
+            {'value': 'hd', 'label': 'hd'},
+        ],
+        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=10,
+        _step=1,
+        precision=0)
+
+
+class XinferenceTextToImageModelCredential(BaseForm, BaseModelCredential):
+    api_base = forms.TextInputField('API 域名', required=True)
+    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_base', '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', ''))}
+
+    def get_model_params_setting_form(self, model_name):
+        return XinferenceTTIModelParams()

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

@@ -0,0 +1,26 @@
+from typing import Dict
+
+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 XinferenceImage(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)
+        return XinferenceImage(
+            model_name=model_name,
+            openai_api_base=model_credential.get('api_base'),
+            openai_api_key=model_credential.get('api_key'),
+            # stream_options={"include_usage": True},
+            streaming=True,
+            **optional_params,
+        )

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

@@ -0,0 +1,58 @@
+from typing import Dict
+
+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_tti import BaseTextToImage
+
+
+def custom_get_token_ids(text: str):
+    tokenizer = TokenizerManage.get_tokenizer()
+    return tokenizer.encode(text)
+
+
+class XinferenceTextToImage(MaxKBBaseModel, BaseTextToImage):
+    api_base: str
+    api_key: str
+    model: str
+    params: dict
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self.api_key = kwargs.get('api_key')
+        self.api_base = kwargs.get('api_base')
+        self.model = kwargs.get('model')
+        self.params = kwargs.get('params')
+
+    @staticmethod
+    def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs):
+        optional_params = {'params': {'size': '1024x1024', 'quality': 'standard', 'n': 1}}
+        for key, value in model_kwargs.items():
+            if key not in ['model_id', 'use_local', 'streaming']:
+                optional_params['params'][key] = value
+        return XinferenceTextToImage(
+            model=model_name,
+            api_base=model_credential.get('api_base'),
+            api_key=model_credential.get('api_key'),
+            **optional_params,
+        )
+
+    def is_cache_model(self):
+        return False
+
+    def check_auth(self):
+        chat = OpenAI(api_key=self.api_key, base_url=self.api_base)
+        response_list = chat.models.with_raw_response.list()
+
+        # self.generate_image('生成一个小猫图片')
+
+    def generate_image(self, prompt: str, negative_prompt: str = None):
+        chat = OpenAI(api_key=self.api_key, base_url=self.api_base)
+        res = chat.images.generate(model=self.model, prompt=prompt, **self.params)
+        file_urls = []
+        for content in res.data:
+            url = content.url
+            file_urls.append(url)
+
+        return file_urls