arm PR104.x: mixed-quant row sharding arm series (PR104.0 roofline / PR104.1 spike / PR104.2 production) - #113

Open

ddvnguyen wants to merge 8 commits into

fork/pr104-mixed-quant-arm

Commits on Sep 9, 2026

docs: add PR104.x mixed-quant row sharding arm series spec
Hydra Engineering
committed
543f5a2

Commits on Sep 10, 2026

docs: PR104.0 roofline measured and GO decision
Hydra Engineering
committed
3614045
spike: PR104.1 env-toggle mixed-quant row sharding foundation
Hydra Engineering
committed
a5143fd
docs: PR104.2 storage validated, compute blocked — progress report
Hydra Engineering
committed
7db9e64
cuda: harden split-buffer compute path
Hydra Engineering
committed
40ea569
docs: PR104.2 compute crash root-caused and fixed, 9B validated; 27B needs per-op split dispatch
Hydra Engineering
committed
23e204b
cuda: per-op row-split MUL_MAT dispatch for mixed-quant split buffers
Hydra Engineering
committed
a38976d
docs: PR104.2 per-op row-split MUL_MAT lands; 27B mixed-quant fits and runs correct
Hydra Engineering
committed
3915308