All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Added the knowledge pack for GenieFramework org. (alias
:genie
).
- Updated the knowledge pack for core Julia docs (alias
:julia
).
Both were made possible by Splendidbug!
- Fixed a bug in
aihelp()
wherererank=false
andsearch=false
would not block these steps if they were previously enabled.
Added (thanks to Splendidbug!)
- New knowledge packs created by Splendidbug using
DocsScraper.jl
(to be registered soon, created as part of Google Summer of Code). - Knowledge packs for JuliaData org (DataFrames.jl), Plots org, and SciML org.
- Refreshed knowledge packs for Makie org and Tidier org.
- Golden Q&A sets for new packages to evaluate performance.
- Changed default chunk size to 384 based on evaluation results.
- Increased compatibility for PromptingTools to v0.50, enabling the use of the latest chat models.
- Changed const-Ref variables to typed globals to prevent issues encountered in PromptingTools.
- Legacy pack "juliaextra".
- Issues related to const-Ref variables by switching to typed globals.
- Fixed
:gold
pipeline inconsistently updating the embedding dimension
- The new knowledge packs are currently available only for the OpenAI Text embedding Large 3 model.
- Latest chat models can now be used, but embedding models are restricted for prebuilt knowledge packs.
- Increased compat for PromptingTools to v0.37.1 to include bug fixes.
- Increased compat for PromptingTools to v0.37
- Fixed a bug in
build_index
where imports were missing and keywords were not passed properly in all scenarios.
- (Preliminary) Knowledge packs available for Julia docs (
:julia
), Tidier ecosystem (:tidier
), and Makie ecosystem (:makie
). Load withload_index!(:julia)
or several withload_index!([:julia, :tidier])
. - Preferences.jl-based persistence for chat model, embedding model, embedding dimension, and which knowledge packs to load on start. See
AIHelpMe.PREFERENCES
for more details. - Precompilation statements to improve TTFX.
- First Q&A evaluation dataset in folder
evaluations/
.
- Bumped up PromptingTools to v0.21 (brings new RAG capabilities, pretty-printing, etc.)
- Changed default model to be GPT-4 Turbo to improve the answer quality (you can quickly change to "gpt3t" if you want something simple)
- Documentation moved to Vitepress
- Fixed wrong initiation of
CONV_HISTORY
and other globals that led to UndefVarError. Moved several globals toconst Ref{}
pattern to ensure type stability, but it means that from now it always needs to be dereferenced with[]
(eg,MAIN_INDEX[]
instead ofMAIN_INDEX
).