tensor4all temporary fork. The
t4a-cubek*crates published from this repository are temporary tensor4all fork packages used bytenferro-rswhile required CubeK/CubeCL patches are being upstreamed. They are not a replacement for upstream CubeK, and the intent is to return to upstream package names once the needed patches have landed.
CubeK: high-performance multi-platform kernels in CubeCL
| Algorithms | Variants |
|---|---|
| Random | bernoulli normal uniform |
| Quantization | symmetric per-block per-tensor q2 q4 q8 fp4 |
| Reduction | mean sum prod max min arg[max|min] per-cube per-plane |
| Matmul | mma unit tma multi-stage specialization ordered multi-rows |
| Convolution | mma unit tma multi-stage im2col |
| Attention | mma unit multi-rows |
If you want to contribute new kernels, please read the GUIDE.md.
The full testing guide — suites, CUBE_TEST_MODE, failure-message format, and
filter syntax — lives in cubek-test-utils.