Skip to content

armgit / burr

Public Imported from GitHub
0

Build applications that make decisions (chatbots, agents, simulations, etc...). Monitor, trace, persist, and execute on your own infrastructure.

Adds support for lc documents & similar

Stefan Krawczyk4c68bdebd637
diff --git a/burr/tracking/common/models.py b/burr/tracking/common/models.py
index 75a275a3..b6f37b8d 100644
--- a/burr/tracking/common/models.py
+++ b/burr/tracking/common/models.py
@@ -20,9 +20,13 @@ except ImportError as e:
try:
# try to import to serialize Langchain messages
+ from langchain_core import documents as lc_documents
+ from langchain_core import load as lc_serde
from langchain_core import messages as lc_messages
except ImportError:
lc_messages = None
+ lc_documents = None
+ lc_serde = None
class IdentifyingModel(pydantic.BaseModel):
@@ -124,13 +128,22 @@ class BeginEntryModel(IdentifyingModel):
return _serialize_object(_filter_inputs(inputs))
-def _serialize_object(d: object) -> Union[dict, list, object]:
+def _serialize_object(d: object) -> Union[dict, list, object, str]:
if isinstance(d, list):
return [_serialize_object(x) for x in d]
elif isinstance(d, dict):
return {k: _serialize_object(v) for k, v in d.items()}
elif lc_messages is not None and isinstance(d, lc_messages.BaseMessage):
return lc_messages.message_to_dict(d)
+ elif lc_documents is not None and isinstance(d, lc_documents.Document):
+ if d.is_lc_serializable():
+ return lc_serde.dumpd(d)
+ else:
+ # d.page_content # hack because not all documents are serializable
+ return d.page_content
+ elif hasattr(d, "to_document"):
+ # langchain can have things that look like a document but aren't...
+ return _serialize_object(d.to_document())
elif hasattr(d, "model_dump"): # generic pydantic object
return d.model_dump()
elif hasattr(d, "to_json"):