feat: Volcanic Engine Image Model

Author: liuruibin

apps/application/flow/step_node/image_understand_step_node/impl/base_image_understand_node.py

@@ -12,6 +12,7 @@
 from application.flow.step_node.image_understand_step_node.i_image_understand_node import IImageUnderstandNode
 from dataset.models import File
 from setting.models_provider.tools import get_model_instance_by_model_user_id
+from imghdr import what
 
 
 def _write_context(node_variable: Dict, workflow_variable: Dict, node: INode, workflow, answer: str):
@@ -59,8 +60,9 @@ def write_context(node_variable: Dict, workflow_variable: Dict, node: INode, wor
 
 def file_id_to_base64(file_id: str):
     file = QuerySet(File).filter(id=file_id).first()
-    base64_image = base64.b64encode(file.get_byte()).decode("utf-8")
-    return base64_image
+    file_bytes = file.get_byte()
+    base64_image = base64.b64encode(file_bytes).decode("utf-8")
+    return [base64_image, what(None, file_bytes)]
 
 
 class BaseImageUnderstandNode(IImageUnderstandNode):
@@ -152,7 +154,7 @@ def generate_history_human_message(self, chat_record):
                 return HumanMessage(
                     content=[
                         {'type': 'text', 'text': data['question']},
-                        *[{'type': 'image_url', 'image_url': {'url': f'data:image/jpeg;base64,{base64_image}'}} for
+                        *[{'type': 'image_url', 'image_url': {'url': f'data:image/{base64_image[1]};base64,{base64_image[0]}'}} for
                           base64_image in image_base64_list]
                     ])
         return HumanMessage(content=chat_record.problem_text)
@@ -167,8 +169,10 @@ def generate_message_list(self, image_model, system: str, prompt: str, history_m
             for img in image:
                 file_id = img['file_id']
                 file = QuerySet(File).filter(id=file_id).first()
-                base64_image = base64.b64encode(file.get_byte()).decode("utf-8")
-                images.append({'type': 'image_url', 'image_url': {'url': f'data:image/jpeg;base64,{base64_image}'}})
+                image_bytes = file.get_byte()
+                base64_image = base64.b64encode(image_bytes).decode("utf-8")
+                image_format = what(None, image_bytes)
+                images.append({'type': 'image_url', 'image_url': {'url': f'data:image/{image_format};base64,{base64_image}'}})
             messages = [HumanMessage(
                 content=[
                     {'type': 'text', 'text': self.workflow_manage.generate_prompt(prompt)},

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

@@ -0,0 +1,63 @@
+# 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 VolcanicEngineImageModelParams(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=0.95,
+                                    _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=1024,
+        _min=1,
+        _max=100000,
+        _step=1,
+        precision=0)
+
+class VolcanicEngineImageModelCredential(BaseForm, BaseModelCredential):
+    api_key = forms.PasswordInputField('API Key', required=True)
+    api_base = forms.TextInputField('API 域名', 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', 'api_base']:
+            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 VolcanicEngineImageModelParams()

apps/setting/models_provider/impl/volcanic_engine_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 VolcanicEngineImage(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 VolcanicEngineImage(
+            model_name=model_name,
+            openai_api_key=model_credential.get('api_key'),
+            openai_api_base=model_credential.get('api_base'),
+            # stream_options={"include_usage": True},
+            streaming=True,
+            **optional_params,
+        )

apps/setting/models_provider/impl/volcanic_engine_model_provider/volcanic_engine_model_provider.py

@@ -14,7 +14,10 @@
 from setting.models_provider.impl.openai_model_provider.credential.embedding import OpenAIEmbeddingCredential
 from setting.models_provider.impl.openai_model_provider.credential.llm import OpenAILLMModelCredential
 from setting.models_provider.impl.openai_model_provider.model.embedding import OpenAIEmbeddingModel
+from setting.models_provider.impl.volcanic_engine_model_provider.credential.image import \
+    VolcanicEngineImageModelCredential
 from setting.models_provider.impl.volcanic_engine_model_provider.credential.tts import VolcanicEngineTTSModelCredential
+from setting.models_provider.impl.volcanic_engine_model_provider.model.image import VolcanicEngineImage
 from setting.models_provider.impl.volcanic_engine_model_provider.model.llm import VolcanicEngineChatModel
 from setting.models_provider.impl.volcanic_engine_model_provider.credential.stt import VolcanicEngineSTTModelCredential
 from setting.models_provider.impl.volcanic_engine_model_provider.model.stt import VolcanicEngineSpeechToText
@@ -25,13 +28,19 @@
 volcanic_engine_llm_model_credential = OpenAILLMModelCredential()
 volcanic_engine_stt_model_credential = VolcanicEngineSTTModelCredential()
 volcanic_engine_tts_model_credential = VolcanicEngineTTSModelCredential()
+volcanic_engine_image_model_credential = VolcanicEngineImageModelCredential()
 
 model_info_list = [
     ModelInfo('ep-xxxxxxxxxx-yyyy',
               '用户前往火山方舟的模型推理页面创建推理接入点，这里需要输入ep-xxxxxxxxxx-yyyy进行调用',
               ModelTypeConst.LLM,
               volcanic_engine_llm_model_credential, VolcanicEngineChatModel
               ),
+    ModelInfo('ep-xxxxxxxxxx-yyyy',
+              '用户前往火山方舟的模型推理页面创建推理接入点，这里需要输入ep-xxxxxxxxxx-yyyy进行调用',
+              ModelTypeConst.IMAGE,
+              volcanic_engine_image_model_credential, VolcanicEngineImage
+              ),
     ModelInfo('asr',
               '',
               ModelTypeConst.STT,
@@ -51,8 +60,13 @@
               ModelTypeConst.EMBEDDING, open_ai_embedding_credential,
               OpenAIEmbeddingModel)]
 
-model_info_manage = ModelInfoManage.builder().append_model_info_list(model_info_list).append_default_model_info(
-    model_info_list[0]).build()
+model_info_manage = (
+    ModelInfoManage.builder()
+    .append_model_info_list(model_info_list)
+    .append_default_model_info(model_info_list[0])
+    .append_default_model_info(model_info_list[1])
+    .build()
+)
 
 
 class VolcanicEngineModelProvider(IModelProvider):