Skip to content

Agent Zero AI framework

fix embedding cache

frdelb7f1daacfa05
diff --git a/models.py b/models.py
index 5e4317158..41a70934b 100644
--- a/models.py
+++ b/models.py
@@ -1,4 +1,5 @@
from enum import Enum
+import logging
import os
from typing import (
Any,
@@ -14,6 +15,7 @@ from typing import (
from litellm import completion, acompletion, embedding
import litellm
+
from python.helpers import dotenv
from python.helpers.dotenv import load_dotenv
from python.helpers.rate_limiter import RateLimiter
@@ -35,10 +37,14 @@ from langchain.embeddings.base import Embeddings
from sentence_transformers import SentenceTransformer
-# disable extra logging
+# disable extra logging, must be done repeatedly, otherwise browser-use will turn it back on for some reason
def turn_off_logging():
os.environ["LITELLM_LOG"] = "ERROR" # only errors
litellm.suppress_debug_info = True
+ # Silence **all** LiteLLM sub-loggers (utils, cost_calculator…)
+ for name in logging.Logger.manager.loggerDict:
+ if name.lower().startswith("litellm"):
+ logging.getLogger(name).setLevel(logging.ERROR)
# init
@@ -392,13 +398,13 @@ class LiteLLMEmbeddingWrapper(Embeddings):
class LocalSentenceTransformerWrapper(Embeddings):
"""Local wrapper for sentence-transformers models to avoid HuggingFace API calls"""
- def __init__(self, model_name: str, **kwargs: Any):
+ def __init__(self, provider: str, model: str, **kwargs: Any):
# Remove the "sentence-transformers/" prefix if present
- if model_name.startswith("sentence-transformers/"):
- model_name = model_name[len("sentence-transformers/") :]
+ if model.startswith("sentence-transformers/"):
+ model = model[len("sentence-transformers/") :]
- self.model = SentenceTransformer(model_name, **kwargs)
- self.model_name = model_name
+ self.model = SentenceTransformer(model, **kwargs)
+ self.model_name = model
def embed_documents(self, texts: List[str]) -> List[List[float]]:
embeddings = self.model.encode(texts, convert_to_tensor=False) # type: ignore
@@ -450,14 +456,14 @@ def get_litellm_embedding(model_name: str, provider: str, **kwargs: Any):
# Check if this is a local sentence-transformers model
if provider == "huggingface" and model_name.startswith("sentence-transformers/"):
# Use local sentence-transformers instead of LiteLLM for local models
- return LocalSentenceTransformerWrapper(model_name=model_name, **kwargs)
+ return LocalSentenceTransformerWrapper(provider=provider, model=model_name, **kwargs)
configure_litellm_environment()
# Use original provider name for API key lookup, fallback to mapped provider name
api_key = kwargs.pop("api_key", None) or get_api_key(provider)
# litellm will pick up base_url from env. We just need to control the api_key.
- base_url = dotenv.get_dotenv_value(f"{provider.upper()}_BASE_URL")
+ # base_url = dotenv.get_dotenv_value(f"{provider.upper()}_BASE_URL")
# If a base_url is set, ensure api_key is not passed to litellm
# > remove, this can be handled by api_key=None
diff --git a/python/helpers/document_query.py b/python/helpers/document_query.py
index 1cfd41845..df7fae415 100644
--- a/python/helpers/document_query.py
+++ b/python/helpers/document_query.py
@@ -20,21 +20,10 @@ from langchain_community.document_transformers import MarkdownifyTransformer
from langchain_community.document_loaders.parsers.images import TesseractBlobParser
from langchain_core.documents import Document
-from langchain.prompts import ChatPromptTemplate
from langchain.schema import SystemMessage, HumanMessage
-from langchain.storage import LocalFileStore, InMemoryStore
-from langchain.embeddings import CacheBackedEmbeddings
-
-from langchain_community.vectorstores import FAISS
-import faiss
-from langchain_community.docstore.in_memory import InMemoryDocstore
-from langchain_community.vectorstores.utils import (
- DistanceStrategy,
-)
-from langchain_core.embeddings import Embeddings
from python.helpers.print_style import PrintStyle
-from python.helpers import files
+from python.helpers import files, errors
from agent import Agent
from langchain.text_splitter import RecursiveCharacterTextSplitter
@@ -106,7 +95,7 @@ class DocumentQueryStore:
return normalized
def init_vector_db(self):
- return VectorDB(self.agent)
+ return VectorDB(self.agent, cache=True)
async def add_document(
self, text: str, document_uri: str, metadata: dict | None = None
@@ -168,7 +157,8 @@ class DocumentQueryStore:
)
return True, ids
except Exception as e:
- PrintStyle.error(f"Error adding document '{document_uri}': {str(e)}")
+ err_text = errors.format_error(e)
+ PrintStyle.error(f"Error adding document '{document_uri}': {err_text}")
return False, []
async def get_document(self, document_uri: str) -> Optional[Document]:
diff --git a/python/helpers/vector_db.py b/python/helpers/vector_db.py
index 128cccfbf..c37a742b0 100644
--- a/python/helpers/vector_db.py
+++ b/python/helpers/vector_db.py
@@ -36,12 +36,14 @@ class VectorDB:
_cached_embeddings: dict[str, CacheBackedEmbeddings] = {}
@staticmethod
- def _get_embeddings(agent: Agent):
+ def _get_embeddings(agent: Agent, cache: bool = True):
model = agent.get_embedding_model()
+ if not cache:
+ return model # return raw embeddings if cache is False
namespace = getattr(
model,
- "model",
- getattr(model, "model_name", "default"),
+ "model_name",
+ "default",
)
if namespace not in VectorDB._cached_embeddings:
store = InMemoryByteStore()
@@ -54,9 +56,10 @@ class VectorDB:
)
return VectorDB._cached_embeddings[namespace]
- def __init__(self, agent: Agent):
+ def __init__(self, agent: Agent, cache: bool = True):
self.agent = agent
- self.embeddings = self._get_embeddings(agent)
+ self.cache = cache # store cache preference
+ self.embeddings = self._get_embeddings(agent, cache=cache)
self.index = faiss.IndexFlatIP(len(self.embeddings.embed_query("example")))
self.db = MyFaiss(