- Move from OpenAI to Mistral Embeddings

- Move embedding model settings from tenant to catalog
- BUG: error processing configuration for chunking patterns in HTML_PROCESSOR
- Removed eveai_chat from docker-files and nginx configuration, as it is now obsolete
- BUG: error in Library Operations when creating a new default RAG library
- BUG: Added public type in migration scripts
- Removed SocketIO from all code and requirements.txt
This commit is contained in:
Josako
2025-02-25 11:17:19 +01:00
parent c037d4135e
commit 55a89c11bb
34 changed files with 457 additions and 444 deletions

View File

@@ -10,7 +10,7 @@ from common.extensions import db
from common.models.document import Document, DocumentVersion, Catalog, Retriever
from common.models.user import Tenant
from common.utils.datetime_utils import get_date_in_timezone
from common.utils.model_utils import get_model_variables
from common.utils.model_utils import get_embedding_model_and_class
from .base import BaseRetriever
from .registry import RetrieverRegistry
@@ -25,10 +25,10 @@ class StandardRAGRetriever(BaseRetriever):
retriever = Retriever.query.get_or_404(retriever_id)
self.catalog_id = retriever.catalog_id
self.tenant_id = tenant_id
self.similarity_threshold = retriever.configuration.get('es_similarity_threshold', 0.3)
self.k = retriever.configuration.get('es_k', 8)
self.tuning = retriever.tuning
self.model_variables = get_model_variables(self.tenant_id)
self.log_tuning("Standard RAG retriever initialized")
@@ -161,8 +161,9 @@ class StandardRAGRetriever(BaseRetriever):
def _get_query_embedding(self, query: str):
"""Get embedding for the query text"""
embedding_model = self.model_variables.embedding_model
return embedding_model.embed_query(query)
catalog = Catalog.query.get_or_404(self.catalog_id)
embedding_model, embedding_model_class = get_embedding_model_and_class(self.tenant_id, self.catalog_id,
catalog.embedding_model)
# Register the retriever type