Damien Lejeune
|
63dcefffc3
|
WIP: v4 tile distribution working
|
2026-02-11 07:51:04 +00:00 |
|
Damien Lejeune
|
7c728adb57
|
Add V4: remove gemm pipeline, combine gemm/normalization
|
2026-02-10 10:39:49 +00:00 |
|
Damien Lejeune
|
6c45f722e7
|
Compute GEMM and normalize in one pass: MHV v3
|
2026-02-10 10:35:10 +00:00 |
|
Damien Lejeune
|
0766752704
|
Refactoring the normalization operation
|
2026-02-09 13:55:54 +00:00 |
|
Damien Lejeune
|
ec1e8ec58e
|
Add benchmark example
|
2026-02-06 14:55:13 +00:00 |
|
Damien Lejeune
|
e7ebd6c288
|
Readd naive normalization in mhc v3
|
2026-02-06 10:41:47 +00:00 |
|
Damien Lejeune
|
053aed9402
|
MHC V3 with gemm pipeline
|
2026-02-05 17:11:09 +00:00 |
|
Damien Lejeune
|
43a5678fdf
|
WIP: MHC v3
|
2026-02-05 13:04:18 +00:00 |
|
Damien Lejeune
|
6ea40157f1
|
Add last steps: activations functions
|
2026-02-02 02:55:17 -05:00 |
|
Damien Lejeune
|
da895cdd88
|
Tile on the C dimensions to support large C
|
2026-01-29 08:00:34 -05:00 |
|
Damien Lejeune
|
c83b1c482b
|
Remove hard coded lds size
|
2026-01-29 05:24:19 -05:00 |
|
Damien Lejeune
|
b83c07748c
|
WIP: arbitrary batch dim
|
2026-01-28 06:00:10 -05:00 |
|
Damien Lejeune
|
389639fe34
|
WIP: add naive version + block gemm version + tests & reference
|
2026-01-27 08:22:36 -05:00 |
|
Damien Lejeune
|
927d121cb8
|
WIP: project setup
|
2026-01-22 11:36:29 -05:00 |
|