gmakstutis/openrag
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

search embeding model support for watsonx models

Browse source
This commit is contained in:
phact 2025-12-02 11:58:01 -05:00
parent 0fc0be855c
commit 0a2a991519
1 changed files with 4 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

11
src/services/search_service.py
View file

@ -1,7 +1,7 @@
import copy import copy
from typing import Any, Dict from typing import Any, Dict
from agentd.tool_decorator import tool from agentd.tool_decorator import tool
from config.settings import EMBED_MODEL, clients, INDEX_NAME, get_embedding_model from config.settings import EMBED_MODEL, clients, INDEX_NAME, get_embedding_model, WATSONX_EMBEDDING_DIMENSIONS
from auth_context import get_auth_context from auth_context import get_auth_context
from utils.logging_config import get_logger from utils.logging_config import get_logger
@ -155,18 +155,15 @@ class SearchService:
if not any(model_name.startswith(prefix + "/") for prefix in ["openai", "ollama", "watsonx", "anthropic"]): if not any(model_name.startswith(prefix + "/") for prefix in ["openai", "ollama", "watsonx", "anthropic"]):
# Detect provider from model name characteristics: # Detect provider from model name characteristics:
# - Ollama: contains ":" (e.g., "nomic-embed-text:latest") # - Ollama: contains ":" (e.g., "nomic-embed-text:latest")
# - WatsonX: starts with "ibm/" or known third-party models # - WatsonX: check against known IBM embedding models
# - OpenAI: everything else (no prefix needed) # - OpenAI: everything else (no prefix needed)
if ":" in model_name: if ":" in model_name:
# Ollama models use tags with colons # Ollama models use tags with colons
formatted_model = f"ollama/{model_name}" formatted_model = f"ollama/{model_name}"
logger.debug(f"Formatted Ollama model: {model_name} -> {formatted_model}") logger.debug(f"Formatted Ollama model: {model_name} -> {formatted_model}")
elif model_name.startswith("ibm/") or model_name in [ elif model_name in WATSONX_EMBEDDING_DIMENSIONS:
"intfloat/multilingual-e5-large", # WatsonX embedding models - use hardcoded list from settings
"sentence-transformers/all-minilm-l6-v2"
]:
# WatsonX embedding models
formatted_model = f"watsonx/{model_name}" formatted_model = f"watsonx/{model_name}"
logger.debug(f"Formatted WatsonX model: {model_name} -> {formatted_model}") logger.debug(f"Formatted WatsonX model: {model_name} -> {formatted_model}")
# else: OpenAI models don't need a prefix # else: OpenAI models don't need a prefix