From 5c93e5d7a89ac2417bb583d6869fe08c16f57cd7 Mon Sep 17 00:00:00 2001
From: furyhawk <furyx@hotmail.com>
Date: Sun, 14 Jun 2026 13:33:34 +0800
Subject: [PATCH] feat: Add session commit in database operations and update
 embedding configuration

---
 backend/core/database.py          | 1 +
 backend/services/rag/config.py    | 8 ++++++--
 backend/worker/tasks/rag_tasks.py | 2 ++
 docker-compose.yml                | 2 +-
 4 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/backend/core/database.py b/backend/core/database.py
index 7a36916..5a4a2ee 100644
--- a/backend/core/database.py
+++ b/backend/core/database.py
@@ -99,6 +99,7 @@ def open_session() -> AsyncSession:
 
         async with open_session() as session:
             await session.execute(...)
+            await session.commit()  # required — no auto-commit
 
     The session is automatically closed on exit of the ``async with`` block.
     """
diff --git a/backend/services/rag/config.py b/backend/services/rag/config.py
index 50f28c8..36e8cb1 100644
--- a/backend/services/rag/config.py
+++ b/backend/services/rag/config.py
@@ -98,14 +98,18 @@ EMBEDDING_DIMENSIONS: dict[str, int] = {
     "bge-small-en-v1.5": 384,
     "bge-base-en-v1.5": 768,
     "bge-large-en-v1.5": 1024,
+    # llama.cpp / local models (Gemma, Llama, etc.)
+    "gemma-4-12b-it": 3840,
+    "gemma-4-12b": 3840,
+    "unsloth/gemma-4-12B-it-qat-GGUF:UD-Q4_K_XL": 3840,
 }
 
 
 class EmbeddingsConfig(BaseModel):
     """Embeddings configuration. Dimension is auto-derived from model name."""
 
-    model: str = "text-embedding-3-small"
-    dim: int = 1536
+    model: str = "unsloth/gemma-4-12B-it-qat-GGUF:UD-Q4_K_XL"
+    dim: int = 3840
 
     @model_validator(mode="after")
     def set_dim_from_model(self) -> "EmbeddingsConfig":
diff --git a/backend/worker/tasks/rag_tasks.py b/backend/worker/tasks/rag_tasks.py
index c997268..db8735e 100644
--- a/backend/worker/tasks/rag_tasks.py
+++ b/backend/worker/tasks/rag_tasks.py
@@ -86,6 +86,7 @@ async def ingest_document_task(
                     doc_id=rag_document_id,
                     error_message=result.error_message or "Unknown error",
                 )
+            await session.commit()
 
         logger.info(
             "[WORKER] Ingestion complete: doc=%s, status=%s",
@@ -104,6 +105,7 @@ async def ingest_document_task(
                 doc_id=rag_document_id,
                 error_message=str(exc),
             )
+            await session.commit()
         return {"status": "error", "message": str(exc)}
 
 
diff --git a/docker-compose.yml b/docker-compose.yml
index 6d9b9a6..0666215 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -42,7 +42,7 @@ services:
       - MAX_UPLOAD_SIZE_MB=${MAX_UPLOAD_SIZE_MB:-50}
       # Embeddings — defaults to LLM_BASE_URL; set to a server that supports
       # the OpenAI-compatible embeddings endpoint (e.g. Ollama, OpenAI).
-      - EMBEDDING_BASE_URL=${EMBEDDING_BASE_URL:-}
+      - EMBEDDING_BASE_URL=${EMBEDDING_BASE_URL:-http://host.containers.internal:8011/v1}
       - EMBEDDING_API_KEY=${EMBEDDING_API_KEY:-}
 
     restart: unless-stopped