refactor: update model provider

libs/admin-api-lib/src/admin_api_lib/dependency_container.py

@@ -9,7 +9,6 @@
     Singleton,
 )
 from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain_community.llms import Ollama, VLLMOpenAI
 from langfuse import Langfuse
 
 from admin_api_lib.extractor_api_client.openapi_client.api.extractor_api import (
@@ -61,7 +60,7 @@
     Configuration as RagConfiguration,
 )
 from rag_core_lib.impl.langfuse_manager.langfuse_manager import LangfuseManager
-from rag_core_lib.impl.llms.llm_factory import llm_provider
+from rag_core_lib.impl.llms.llm_factory import chat_model_provider
 from rag_core_lib.impl.settings.langfuse_settings import LangfuseSettings
 from rag_core_lib.impl.settings.ollama_llm_settings import OllamaSettings
 from rag_core_lib.impl.settings.rag_class_types_settings import RAGClassTypeSettings
@@ -108,8 +107,8 @@ class DependencyContainer(DeclarativeContainer):
 
     large_language_model = Selector(
         class_selector_config.llm_type,
-        ollama=Singleton(llm_provider, ollama_settings, Ollama),
-        stackit=Singleton(llm_provider, stackit_vllm_settings, VLLMOpenAI),
+        ollama=Singleton(chat_model_provider, ollama_settings, "ollama"),
+        stackit=Singleton(chat_model_provider, stackit_vllm_settings, "openai"),
     )
 
     summary_text_splitter = Singleton(RecursiveCharacterTextSplitter)(

libs/admin-api-lib/src/admin_api_lib/impl/summarizer/langchain_summarizer.py

@@ -77,15 +77,17 @@ async def ainvoke(self, query: SummarizerInput, config: Optional[RunnableConfig]
         for langchain_document in langchain_documents:
             async with self._semaphore:
                 try:
-                    outputs.append(
-                        await self._create_chain().ainvoke({"text": langchain_document.page_content}, config)
-                    )
+                    result = await self._create_chain().ainvoke({"text": langchain_document.page_content}, config)
+                    # Extract content from AIMessage if it's not already a string
+                    content = result.content if hasattr(result, "content") else str(result)
+                    outputs.append(content)
                 except Exception as e:
                     logger.error("Error in summarizing langchain doc: %s %s", e, traceback.format_exc())
                     config["tries_remaining"] = tries_remaining - 1
-                    outputs.append(
-                        await self._create_chain().ainvoke({"text": langchain_document.page_content}, config)
-                    )
+                    result = await self._create_chain().ainvoke({"text": langchain_document.page_content}, config)
+                    # Extract content from AIMessage if it's not already a string
+                    content = result.content if hasattr(result, "content") else str(result)
+                    outputs.append(content)
 
         if len(outputs) == 1:
             return outputs[0]

libs/admin-api-lib/src/admin_api_lib/prompt_templates/summarize_prompt.py

@@ -1,8 +1,14 @@
 """Module for the summarize prompt template."""
 
-SUMMARIZE_PROMPT = """<|begin_of_text|><|start_header_id|>system<|end_header_id|>
+from langchain_core.prompts import ChatPromptTemplate, SystemMessagePromptTemplate, HumanMessagePromptTemplate
 
-Fasse den folgenden Text zusammen. Die Ausgabe soll auf Deutsch sein. Gebe nur die deutsche Zusammenfassung aus, keinen zusätzlichen Text. Halte dich möglichst kurz mit deiner Antwort.<|eot_id|><|start_header_id|>user<|end_header_id|>
-
-{text}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
-"""
+SUMMARIZE_PROMPT = ChatPromptTemplate.from_messages(
+    [
+        SystemMessagePromptTemplate.from_template(
+            "Fasse den folgenden Text zusammen. Die Ausgabe soll auf Deutsch sein. "
+            "Gebe nur die deutsche Zusammenfassung aus, keinen zusätzlichen Text. "
+            "Halte dich möglichst kurz mit deiner Antwort."
+        ),
+        HumanMessagePromptTemplate.from_template("{text}"),
+    ]
+)