Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions src/memos/api/config.py
Original file line number Diff line number Diff line change
Expand Up @@ -979,6 +979,10 @@ def get_product_default_config() -> dict[str, Any]:
"general_llm": APIConfig.get_memreader_general_llm_config(),
# Image parser LLM (requires vision model)
"image_parser_llm": APIConfig.get_image_parser_llm_config(),
# Preference extractor LLM. Reader falls back to general_llm when unset.
"preference_extractor_llm": APIConfig.get_preference_extractor_llm_config()
if os.getenv("PREFERENCE_EXTRACTOR_MODEL")
else None,
"embedder": APIConfig.get_embedder_config(),
"chunker": {
"backend": "sentence",
Expand Down Expand Up @@ -1108,6 +1112,10 @@ def create_user_config(user_name: str, user_id: str) -> tuple["MOSConfig", "Gene
"general_llm": APIConfig.get_memreader_general_llm_config(),
# Image parser LLM (requires vision model)
"image_parser_llm": APIConfig.get_image_parser_llm_config(),
# Preference extractor LLM. Reader falls back to general_llm when unset.
"preference_extractor_llm": APIConfig.get_preference_extractor_llm_config()
if os.getenv("PREFERENCE_EXTRACTOR_MODEL")
else None,
"embedder": APIConfig.get_embedder_config(),
"chunker": {
"backend": "sentence",
Expand Down
4 changes: 4 additions & 0 deletions src/memos/configs/mem_reader.py
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,10 @@ def parse_datetime(cls, value):
default=None,
description="Vision LLM for image parsing. Falls back to general_llm if not set.",
)
preference_extractor_llm: LLMConfigFactory | None = Field(
default=None,
description="LLM for preference extraction. Falls back to general_llm if not set.",
)
embedder: EmbedderConfigFactory = Field(
..., description="Embedder configuration for the MemReader"
)
Expand Down
4 changes: 2 additions & 2 deletions src/memos/mem_reader/multi_modal_struct.py
Original file line number Diff line number Diff line change
Expand Up @@ -1035,7 +1035,7 @@ def _process_multi_modal_data(
process_preference_fine,
non_file_url_fast_items,
info,
self.general_llm,
getattr(self, "preference_extractor_llm", self.general_llm),
self.embedder,
**kwargs,
)
Expand Down Expand Up @@ -1127,7 +1127,7 @@ def _process_transfer_multi_modal_data(
process_preference_fine,
non_file_url_nodes,
info,
self.general_llm,
getattr(self, "preference_extractor_llm", self.general_llm),
self.embedder,
**kwargs,
)
Expand Down
6 changes: 6 additions & 0 deletions src/memos/mem_reader/simple_struct.py
Original file line number Diff line number Diff line change
Expand Up @@ -184,6 +184,12 @@ def __init__(self, config: SimpleStructMemReaderConfig):
if config.general_llm is not None
else self.llm
)
preference_extractor_llm_config = getattr(config, "preference_extractor_llm", None)
self.preference_extractor_llm = (
LLMFactory.from_config(preference_extractor_llm_config)
if preference_extractor_llm_config is not None
else self.general_llm
)
self.qwen_llm = None
qwen_llm_config = getattr(config, "qwen_llm", None)
if qwen_llm_config:
Expand Down
134 changes: 134 additions & 0 deletions tests/mem_reader/test_preference_extractor_llm_config.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,134 @@
from types import SimpleNamespace
from unittest.mock import MagicMock, patch

from memos.api.config import APIConfig
from memos.chunkers import ChunkerFactory
from memos.embedders.factory import EmbedderFactory
from memos.llms.factory import LLMFactory
from memos.mem_reader.multi_modal_struct import MultiModalStructMemReader
from memos.mem_reader.simple_struct import SimpleStructMemReader
from memos.memories.textual.item import (
SourceMessage,
TextualMemoryItem,
TreeNodeTextualMemoryMetadata,
)


def test_product_default_config_wires_preference_extractor_model(monkeypatch):
monkeypatch.setenv("PREFERENCE_EXTRACTOR_MODEL", "pref-model")
monkeypatch.setenv("PREFERENCE_EXTRACTOR_API_BASE", "https://pref.example/v1")
monkeypatch.setenv("PREFERENCE_EXTRACTOR_API_KEY", "pref-key")

config = APIConfig.get_product_default_config()["mem_reader"]["config"]

pref_config = config["preference_extractor_llm"]
assert pref_config["backend"] == "openai"
assert pref_config["config"]["model_name_or_path"] == "pref-model"
assert pref_config["config"]["api_base"] == "https://pref.example/v1"
assert pref_config["config"]["api_key"] == "pref-key"


def test_product_default_config_leaves_preference_extractor_unset_without_model(monkeypatch):
monkeypatch.delenv("PREFERENCE_EXTRACTOR_MODEL", raising=False)
monkeypatch.delenv("PREFERENCE_EXTRACTOR_API_BASE", raising=False)
monkeypatch.delenv("PREFERENCE_EXTRACTOR_API_KEY", raising=False)

config = APIConfig.get_product_default_config()["mem_reader"]["config"]

assert config["preference_extractor_llm"] is None


def test_simple_reader_uses_configured_preference_extractor_llm():
main_llm = object()
general_llm = object()
preference_llm = object()

config = SimpleNamespace(
llm="main-config",
general_llm="general-config",
preference_extractor_llm="preference-config",
qwen_llm=None,
embedder="embedder-config",
chunker="chunker-config",
remove_prompt_example=False,
)
chunker = MagicMock()
chunker.config.save_rawfile = False

with (
patch.object(
LLMFactory,
"from_config",
side_effect=[main_llm, general_llm, preference_llm],
),
patch.object(EmbedderFactory, "from_config", return_value=MagicMock()),
patch.object(ChunkerFactory, "from_config", return_value=chunker),
):
reader = SimpleStructMemReader(config)

assert reader.llm is main_llm
assert reader.general_llm is general_llm
assert reader.preference_extractor_llm is preference_llm


def test_simple_reader_preference_extractor_llm_falls_back_to_general_llm():
main_llm = object()
general_llm = object()

config = SimpleNamespace(
llm="main-config",
general_llm="general-config",
preference_extractor_llm=None,
qwen_llm=None,
embedder="embedder-config",
chunker="chunker-config",
remove_prompt_example=False,
)
chunker = MagicMock()
chunker.config.save_rawfile = False

with (
patch.object(LLMFactory, "from_config", side_effect=[main_llm, general_llm]),
patch.object(EmbedderFactory, "from_config", return_value=MagicMock()),
patch.object(ChunkerFactory, "from_config", return_value=chunker),
):
reader = SimpleStructMemReader(config)

assert reader.llm is main_llm
assert reader.general_llm is general_llm
assert reader.preference_extractor_llm is general_llm


def test_multimodal_transfer_uses_preference_extractor_llm():
reader = MultiModalStructMemReader.__new__(MultiModalStructMemReader)
reader.general_llm = object()
reader.preference_extractor_llm = object()
reader.embedder = object()
reader.searcher = None
reader.graph_db = None
reader.oss_config = None
reader.skills_dir_config = None
reader.multi_modal_parser = MagicMock()
reader.multi_modal_parser.process_transfer.return_value = []

raw_node = TextualMemoryItem(
memory="User likes concise answers.",
metadata=TreeNodeTextualMemoryMetadata(
user_id="u1",
session_id="s1",
memory_type="LongTermMemory",
sources=[SourceMessage(type="chat", role="user", content="I like concise answers.")],
),
)

with (
patch.object(reader, "_process_string_fine", return_value=[]),
patch.object(reader, "_process_tool_trajectory_fine", return_value=[]),
patch("memos.mem_reader.multi_modal_struct.process_skill_memory_fine", return_value=[]),
patch(
"memos.mem_reader.multi_modal_struct.process_preference_fine", return_value=[]
) as mock_process_pref,
):
reader._process_transfer_multi_modal_data([raw_node])

assert mock_process_pref.call_args.args[2] is reader.preference_extractor_llm
Loading