Feature/41 add multi-llm provider

.env.example

@@ -0,0 +1,66 @@
+###############################################
+############## LLM API SELECTION ##############
+###############################################
+
+# LLM_PROVIDER=openai
+# OPEN_AI_LLM_KEY=
+# OPEN_AI_LLM_MODEL=gpt-4o
+
+# LLM_PROVIDER=gemini
+# GEMINI_API_KEY=
+# GEMINI_LLM_MODEL=gemini-2.0-flash-lite
+
+# LLM_PROVIDER=azure
+# AZURE_OPENAI_LLM_ENDPOINT=
+# AZURE_OPENAI_LLM_KEY=
+# AZURE_OPENAI_LLM_MODEL=
+# AZURE_OPENAI_LLM_API_VERSION=
+
+# LLM_PROVIDER=ollama
+# OLLAMA_LLM_BASE_URL=
+# OLLAMA_LLM_MODEL=
+
+# LLM_PROVIDER=huggingface
+# HUGGING_FACE_LLM_REPO_ID=
+# HUGGING_FACE_LLM_ENDPOINT=
+# HUGGING_FACE_LLM_API_TOKEN=
+
+# LLM_PROVIDER=bedrock
+# AWS_BEDROCK_LLM_ACCESS_KEY_ID=
+# AWS_BEDROCK_LLM_SECRET_ACCESS_KEY=
+# AWS_BEDROCK_LLM_REGION=us-west-2
+# AWS_BEDROCK_LLM_ENDPOINT_URL=https://bedrock.us-west-2.amazonaws.com
+# AWS_BEDROCK_LLM_MODEL=anthropic.claude-3-5-sonnet-20241022-v2:0\
+
+###############################################
+########### Embedding API SElECTION ###########
+###############################################
+# Only used if you are using an LLM that does not natively support embedding (openai or Azure)
+# EMBEDDING_ENGINE='openai'
+# OPEN_AI_KEY=sk-xxxx
+# EMBEDDING_MODEL_PREF='text-embedding-ada-002'
+
+# EMBEDDING_ENGINE='azure'
+# AZURE_OPENAI_ENDPOINT=
+# AZURE_OPENAI_KEY=
+# EMBEDDING_MODEL_PREF='my-embedder-model' # This is the "deployment" on Azure you want to use for embeddings. Not the base model. Valid base model is text-embedding-ada-002
+
+# EMBEDDING_ENGINE='ollama'
+# EMBEDDING_BASE_PATH='http://host.docker.internal:11434'
+# EMBEDDING_MODEL_PREF='nomic-embed-text:latest'
+# EMBEDDING_MODEL_MAX_CHUNK_LENGTH=8192
+
+# EMBEDDING_ENGINE='bedrock'
+# AWS_BEDROCK_EMBEDDING_ACCESS_KEY_ID=
+# AWS_BEDROCK_EMBEDDING_ACCESS_KEY=
+# AWS_BEDROCK_EMBEDDING_REGION=us-west-2
+# AWS_BEDROCK_EMBEDDING_MODEL_PREF=amazon.embedding-embedding-ada-002:0
+
+# EMBEDDING_ENGINE='gemini'
+# GEMINI_EMBEDDING_API_KEY=
+# EMBEDDING_MODEL_PREF='text-embedding-004'
+
+# EMBEDDING_ENGINE='huggingface'
+# HUGGING_FACE_EMBEDDING_REPO_ID=
+# HUGGING_FACE_EMBEDDING_MODEL=
+# HUGGING_FACE_EMBEDDING_API_TOKEN=

llm_utils/llm_factory.py

@@ -1,25 +1,179 @@
 # llm_factory.py
+import os
 from typing import Optional
 
+from dotenv import load_dotenv
 from langchain.llms.base import BaseLanguageModel
-from langchain_openai import ChatOpenAI
+from langchain_aws import ChatBedrockConverse, BedrockEmbeddings
+from langchain_google_genai import ChatGoogleGenerativeAI, GoogleGenerativeAIEmbeddings
+from langchain_huggingface import (
+    ChatHuggingFace,
+    HuggingFaceEndpoint,
+    HuggingFaceEndpointEmbeddings,
+)
+from langchain_ollama import ChatOllama, OllamaEmbeddings
+from langchain_openai import (
+    AzureOpenAIEmbeddings,
+    ChatOpenAI,
+    AzureChatOpenAI,
+    OpenAIEmbeddings,
+)
+from langchain_community.llms.bedrock import Bedrock
 
+# .env 파일 로딩
+load_dotenv()
 
-def get_llm(
-    model_type: str,
-    model_name: Optional[str] = None,
-    openai_api_key: Optional[str] = None,
-    **kwargs,
-) -> BaseLanguageModel:
+
+def get_llm() -> BaseLanguageModel:
     """
-    주어진 model_type과 model_name 등에 따라 적절한 LLM 객체를 생성/반환한다.
+    return chat model interface
     """
-    if model_type == "openai":
-        return ChatOpenAI(
-            model=model_name,
-            api_key=openai_api_key,
-            **kwargs,
+    provider = os.getenv("LLM_PROVIDER")
+
+    if provider is None:
+        raise ValueError("LLM_PROVIDER environment variable is not set.")
+
+    if provider == "openai":
+        return get_llm_openai()
+
+    elif provider == "azure":
+        return get_llm_azure()
+
+    elif provider == "bedrock":
+        return get_llm_bedrock()
+
+    elif provider == "gemini":
+        return get_llm_gemini()
+
+    elif provider == "ollama":
+        return get_llm_ollama()
+
+    elif provider == "huggingface":
+        return get_llm_huggingface()
+
+    else:
+        raise ValueError(f"Invalid LLM API Provider: {provider}")
+
+
+def get_llm_openai() -> BaseLanguageModel:
+    return ChatOpenAI(
+        model=os.getenv("OPEN_MODEL_PREF", "gpt-4o"),
+        api_key=os.getenv("OPEN_AI_KEY"),
+    )
+
+
+def get_llm_azure() -> BaseLanguageModel:
+    return AzureChatOpenAI(
+        api_key=os.getenv("AZURE_OPENAI_LLM_KEY"),
+        azure_endpoint=os.getenv("AZURE_OPENAI_LLM_ENDPOINT"),
+        azure_deployment=os.getenv("AZURE_OPENAI_LLM_MODEL"),  # Deployment name
+        api_version=os.getenv("AZURE_OPENAI_LLM_API_VERSION", "2023-07-01-preview"),
+    )
+
+
+def get_llm_bedrock() -> BaseLanguageModel:
+    return ChatBedrockConverse(
+        model=os.getenv("AWS_BEDROCK_LLM_MODEL"),
+        aws_access_key_id=os.getenv("AWS_BEDROCK_LLM_ACCESS_KEY_ID"),
+        aws_secret_access_key=os.getenv("AWS_BEDROCK_LLM_SECRET_ACCESS_KEY"),
+        region_name=os.getenv("AWS_BEDROCK_LLM_REGION", "us-east-1"),
+    )
+
+
+def get_llm_gemini() -> BaseLanguageModel:
+    return ChatGoogleGenerativeAI(model=os.getenv("GEMINI_LLM_MODEL"))
+
+
+def get_llm_ollama() -> BaseLanguageModel:
+    base_url = os.getenv("OLLAMA_LLM_BASE_URL")
+    if base_url:
+        return ChatOllama(base_url=base_url, model=os.getenv("OLLAMA_LLM_MODEL"))
+    else:
+        return ChatOllama(model=os.getenv("OLLAMA_LLM_MODEL"))
+
+
+def get_llm_huggingface() -> BaseLanguageModel:
+    return ChatHuggingFace(
+        llm=HuggingFaceEndpoint(
+            model=os.getenv("HUGGING_FACE_LLM_MODEL"),
+            repo_id=os.getenv("HUGGING_FACE_LLM_REPO_ID"),
+            task="text-generation",
+            endpoint_url=os.getenv("HUGGING_FACE_LLM_ENDPOINT"),
+            huggingfacehub_api_token=os.getenv("HUGGING_FACE_LLM_API_TOKEN"),
         )
+    )
+
+
+def get_embeddings() -> Optional[BaseLanguageModel]:
+    """
+    return embedding model interface
+    """
+    provider = os.getenv("EMBEDDING_PROVIDER")
+
+    if provider is None:
+        raise ValueError("EMBEDDING_PROVIDER environment variable is not set.")
+
+    if provider == "openai":
+        return get_embeddings_openai()
+
+    elif provider == "bedrock":
+        return get_embeddings_bedrock()
+
+    elif provider == "azure":
+        return get_embeddings_azure()
+
+    elif provider == "gemini":
+        return get_embeddings_gemini()
+
+    elif provider == "ollama":
+        return get_embeddings_ollama()
 
     else:
-        raise ValueError(f"지원하지 않는 model_type: {model_type}")
+        raise ValueError(f"Invalid Embedding API Provider: {provider}")
+
+
+def get_embeddings_openai() -> BaseLanguageModel:
+    return OpenAIEmbeddings(
+        model=os.getenv("OPEN_AI_EMBEDDING_MODEL"),
+        openai_api_key=os.getenv("OPEN_AI_EMBEDDING_KEY"),
+    )
+
+
+def get_embeddings_azure() -> BaseLanguageModel:
+    return AzureOpenAIEmbeddings(
+        api_key=os.getenv("AZURE_OPENAI_EMBEDDING_KEY"),
+        azure_endpoint=os.getenv("AZURE_OPENAI_EMBEDDING_ENDPOINT"),
+        azure_deployment=os.getenv("AZURE_OPENAI_EMBEDDING_MODEL"),
+        api_version=os.getenv("AZURE_OPENAI_EMBEDDING_API_VERSION"),
+    )
+
+
+def get_embeddings_bedrock() -> BaseLanguageModel:
+    return BedrockEmbeddings(
+        model_id=os.getenv("AWS_BEDROCK_EMBEDDING_MODEL"),
+        aws_access_key_id=os.getenv("AWS_BEDROCK_EMBEDDING_ACCESS_KEY_ID"),
+        aws_secret_access_key=os.getenv("AWS_BEDROCK_EMBEDDING_SECRET_ACCESS_KEY"),
+        region_name=os.getenv("AWS_BEDROCK_EMBEDDING_REGION", "us-east-1"),
+    )
+
+
+def get_embeddings_gemini() -> BaseLanguageModel:
+    return GoogleGenerativeAIEmbeddings(
+        model=os.getenv("GEMINI_EMBEDDING_MODEL"),
+        api_key=os.getenv("GEMINI_EMBEDDING_KEY"),
+    )
+
+
+def get_embeddings_ollama() -> BaseLanguageModel:
+    return OllamaEmbeddings(
+        model=os.getenv("OLLAMA_EMBEDDING_MODEL"),
+        base_url=os.getenv("OLLAMA_EMBEDDING_BASE_URL"),
+    )
+
+
+def get_embeddings_huggingface() -> BaseLanguageModel:
+    return HuggingFaceEndpointEmbeddings(
+        model=os.getenv("HUGGING_FACE_EMBEDDING_MODEL"),
+        repo_id=os.getenv("HUGGING_FACE_EMBEDDING_REPO_ID"),
+        huggingfacehub_api_token=os.getenv("HUGGING_FACE_EMBEDDING_API_TOKEN"),
+    )