Repo that demonstrates hows to build langchain data loaders in Flyte to help build production grade langchain apps
- Local execution!
pyflyte run ingest_wf.py ingest
- Remote Execution
pyflyte run --remote ingest_wf.py ingest
- Test remote
- Imagespec to manage dependencies
- Docment TypeTransformer (or pydantic) so that UI can visualize this better
- Better vector store management, so that during the reduction step, vector stores can be loaded lazily so that memory requirements can be lower.
- Visualization through FlyteDecks
- Upadate the langchain serving example to load the FAISS vector store from flyte registry
- add this to UnionML as a default way of writing prod ready Langchain apps