feat: 🎸 ollama

running

.backend_env.example

@@ -10,6 +10,9 @@ GOOGLE_CLOUD_PROJECT=<change-me>
 CELERY_BROKER_URL=redis:https://redis:6379/0
 CELEBRY_BROKER_QUEUE_NAME=quivr-preview.fifo
 
+#LOCAL
+#OLLAMA_API_BASE_URL=http:https://host.docker.internal:11434 # local all in one remove comment to use local llm with Ollama 
+
 
 
 #RESEND

backend/llm/qa_base.py

@@ -7,6 +7,7 @@
 from langchain.chains import ConversationalRetrievalChain, LLMChain
 from langchain.chains.question_answering import load_qa_chain
 from langchain.chat_models import ChatLiteLLM
+from langchain.embeddings.ollama import OllamaEmbeddings
 from langchain.embeddings.openai import OpenAIEmbeddings
 from langchain.llms.base import BaseLLM
 from langchain.prompts.chat import (
@@ -84,8 +85,13 @@ def _determine_callback_array(
  ]
 
  @property
- def embeddings(self) -> OpenAIEmbeddings:
- return OpenAIEmbeddings() # pyright: ignore reportPrivateUsage=none
+ def embeddings(self):
+ if self.brain_settings.ollama_api_base_url:
+ return OllamaEmbeddings(
+ base_url=self.brain_settings.ollama_api_base_url
+ ) # pyright: ignore reportPrivateUsage=none
+ else: 
+ return OpenAIEmbeddings()
 
  supabase_client: Optional[Client] = None
  vector_store: Optional[CustomSupabaseVectorStore] = None

backend/models/settings.py

@@ -1,9 +1,13 @@
-from langchain.embeddings.openai import OpenAIEmbeddings
 from models.databases.supabase.supabase import SupabaseDB
 from pydantic import BaseSettings
 from supabase.client import Client, create_client
 from vectorstore.supabase import SupabaseVectorStore
+from langchain.embeddings.ollama import OllamaEmbeddings
+from langchain.embeddings.openai import OpenAIEmbeddings
+
+from logger import get_logger
 
+logger = get_logger(__name__)
 
 class BrainRateLimiting(BaseSettings):
  max_brain_per_user: int = 5
@@ -40,11 +44,14 @@ def get_supabase_db() -> SupabaseDB:
  return SupabaseDB(supabase_client)
 
 
-def get_embeddings() -> OpenAIEmbeddings:
+def get_embeddings():
  settings = BrainSettings() # pyright: ignore reportPrivateUsage=none
- embeddings = OpenAIEmbeddings(
- openai_api_key=settings.openai_api_key
- ) # pyright: ignore reportPrivateUsage=none
+ if settings.ollama_api_base_url:
+ embeddings = OllamaEmbeddings(
+ base_url=settings.ollama_api_base_url,
+ ) # pyright: ignore reportPrivateUsage=none
+ else:
+ embeddings = OpenAIEmbeddings() # pyright: ignore reportPrivateUsage=none
  return embeddings