Release 0.9.1 - RAG support in the UI and in Apps + streaming fix

Helix has introduced support for RAG (Retrieval-Augmented Generation) in its 0.9 release. Users can now upload documents and perform RAG over them from the homepage. The default Learn mode is now RAG, which is faster and better at retrieving specific facts than fine-tuning. However, fine-tuning can still be used for answering general questions about the uploaded documents. The terms "inference" and "finetune" have been replaced with more user-friendly terms "chat" and "learn".

Additionally, users can now specify RAG and finetune data sources in Helix Apps' `helix.yaml` to customize an assistant with a RAG data source or fine-tuned LLM. This involves running a RAG or finetune session to create a "data source ID", which can be retrieved and placed in a `helix.yaml` file in a GitHub repo. This `rag_source_id` can also be overridden as an API parameter when making an API call. The same can be done with finetune data sources, named `finetune_data_entity_id` in the info panel and specified in the `helix.yaml` as `lora_id`.

Key takeaways:

Helix now supports RAG. Users can upload documents and perform RAG over them from the homepage.
The terms 'inference' and 'finetune' have been replaced with 'chat' and 'learn' for a more user-friendly experience.
The default Learn mode is now RAG as it is faster than fine-tuning and better at retrieving specific facts.
Users can now specify RAG and finetune data sources in Helix Apps' helix.yaml to customize an assistant with a RAG data source or fine tuned LLM.

Release 0.9.1 - RAG support in the UI and in Apps + streaming fix · helixml/helix

Key takeaways:

Comments (0)

Newsletter