ingest funktioniert jetzt mit Jina über HF und ST und mit mistral über GWDG

math-tutor/backend/app/config.py

@@ -19,14 +19,14 @@ def get_embedding_settings() -> EmbeddingSettings:
     if embedding_type == "sentence-transformer":
         return EmbeddingSettings(
             embedding_type=embedding_type,
-            model=os.getenv("SENTENCE_TRANSFORMER_MODEL", "tencent/KaLM-Embedding-Gemma3-12B-2511"),
-            target_dim=int(os.getenv("SENTENCE_TRANSFORMER_TARGET_DIM", "1024")),
+            model=os.getenv("SENTENCE_TRANSFORMER_MODEL", "jinaai/jina-embeddings-v4"),
+            target_dim=int(os.getenv("EMBEDDING_DIM", "512")),
         )
     if embedding_type == "openai-like":
         base_url = os.getenv("OPENAI_BASE_URL")
         api_key = os.getenv("OPENAI_API_KEY")
-        model = os.getenv("OPENAI_EMBED_MODEL", "text-embedding-3-large")
-        model_target_dim = int(os.getenv("OPENAI_EMBED_TARGET_DIM", "1024"))
+        model = os.getenv("OPENAI_EMBED_MODEL", "e5-mistral-7b-instruct")
+        model_target_dim = int(os.getenv("EMBEDDING_DIM", "1024"))
         if not base_url or not api_key:
             raise ValueError("Missing OPENAI_BASE_URL or OPENAI_API_KEY")
         return EmbeddingSettings(

math-tutor/backend/app/deterministic_services/embeddings.py

@@ -7,7 +7,6 @@ from enum import Enum
 import httpx
 from pydantic import BaseModel, Field
 from sentence_transformers import SentenceTransformer
-import torch
 
 
 # -----------------------------

math-tutor/backend/app/deterministic_services/vector_store.py

@@ -12,6 +12,9 @@ from pgvector.psycopg import register_vector
 import yaml
 from pydantic import BaseModel
 
+import app.config
+
+embedding_dim = app.config.get_embedding_settings().target_dim
 
 @dataclass
 class DocRecord:
@@ -116,7 +119,7 @@ def embed_query(embedder: EmbeddingLike, text: str) -> List[float]:
     return embedder.embed_query(text)
 
 
-DDL = """
+DDL = f"""
 CREATE EXTENSION IF NOT EXISTS vector;
 
 CREATE TABLE IF NOT EXISTS docs (
@@ -134,7 +137,7 @@ CREATE TABLE IF NOT EXISTS docs (
   path TEXT NOT NULL,
   markdown TEXT NOT NULL,
 
-  embedding VECTOR(512) NOT NULL
+  embedding VECTOR({embedding_dim}) NOT NULL
 );
 
 CREATE INDEX IF NOT EXISTS docs_embedding_cos_idx