v5.1.1
Fixed
- Time estimates now show BLAS backend caveat when not running on MKL (estimates are calibrated to MKL ILP64 on 48-core Xeon)
- Memory pre-flight check logs active BLAS backend and ILP64 status; warns when >40k samples without ILP64 or when time estimates are uncalibrated
- Fix pip install order in docs: deps first, numpy-mkl second, jamma
--no-depslast
Changed
- High contrast mermaid diagrams across all docs with dark subgraph backgrounds and bright node fills
- Three new diagrams in USER_GUIDE: GWAS pipeline flow, BLAS/eigendecomp dispatch, and memory safety architecture
See CHANGELOG for full details.