feat: Ollama Image understand model

Author: liuruibin

apps/setting/models_provider/impl/ollama_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 OllamaImageModelParams(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 OllamaImageModelCredential(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 OllamaImageModelParams()

apps/setting/models_provider/impl/ollama_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 OllamaImage(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 OllamaImage(
+            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/ollama_model_provider/ollama_model_provider.py

@@ -21,8 +21,10 @@
 from setting.models_provider.base_model_provider import IModelProvider, ModelProvideInfo, ModelInfo, ModelTypeConst, \
     BaseModelCredential, DownModelChunk, DownModelChunkStatus, ValidCode, ModelInfoManage
 from setting.models_provider.impl.ollama_model_provider.credential.embedding import OllamaEmbeddingModelCredential
+from setting.models_provider.impl.ollama_model_provider.credential.image import OllamaImageModelCredential
 from setting.models_provider.impl.ollama_model_provider.credential.llm import OllamaLLMModelCredential
 from setting.models_provider.impl.ollama_model_provider.model.embedding import OllamaEmbedding
+from setting.models_provider.impl.ollama_model_provider.model.image import OllamaImage
 from setting.models_provider.impl.ollama_model_provider.model.llm import OllamaChatModel
 from smartdoc.conf import PROJECT_DIR
 
@@ -133,22 +135,44 @@
         ModelTypeConst.LLM, ollama_llm_model_credential, OllamaChatModel),
 ]
 ollama_embedding_model_credential = OllamaEmbeddingModelCredential()
+ollama_image_model_credential = OllamaImageModelCredential()
 embedding_model_info = [
     ModelInfo(
         'nomic-embed-text',
         '一个具有大令牌上下文窗口的高性能开放嵌入模型。',
         ModelTypeConst.EMBEDDING, ollama_embedding_model_credential, OllamaEmbedding),
 ]
 
-model_info_manage = ModelInfoManage.builder().append_model_info_list(model_info_list).append_model_info_list(
-    embedding_model_info).append_default_model_info(
+image_model_info = [
     ModelInfo(
+        'llava:7b',
+        '',
+        ModelTypeConst.IMAGE, ollama_image_model_credential, OllamaImage),
+    ModelInfo(
+        'llava:13b',
+        '',
+        ModelTypeConst.IMAGE, ollama_image_model_credential, OllamaImage),
+    ModelInfo(
+        'llava:34b',
+        '',
+        ModelTypeConst.IMAGE, ollama_image_model_credential, OllamaImage),
+]
+
+model_info_manage = (
+    ModelInfoManage.builder()
+    .append_model_info_list(model_info_list)
+    .append_model_info_list(embedding_model_info)
+    .append_default_model_info(ModelInfo(
         'phi3',
         'Phi-3 Mini是Microsoft的3.8B参数，轻量级，最先进的开放模型。',
-        ModelTypeConst.LLM, ollama_llm_model_credential, OllamaChatModel)).append_default_model_info(ModelInfo(
-    'nomic-embed-text',
-    '一个具有大令牌上下文窗口的高性能开放嵌入模型。',
-    ModelTypeConst.EMBEDDING, ollama_embedding_model_credential, OllamaEmbedding), ).build()
+        ModelTypeConst.LLM, ollama_llm_model_credential, OllamaChatModel))
+    .append_default_model_info(ModelInfo(
+        'nomic-embed-text',
+        '一个具有大令牌上下文窗口的高性能开放嵌入模型。',
+        ModelTypeConst.EMBEDDING, ollama_embedding_model_credential, OllamaEmbedding), )
+    .append_model_info_list(image_model_info)
+    .build()
+)
 
 
 def get_base_url(url: str):