mfritsche 04d2c6e548 rknpu2: M-bucketing — ceil-to-128 for M>128 instead of next_power_of_two ...

Experiment: next_power_of_two pads prefill M up to the next pow2 (300->512), wasting matmul work. Switch to ceil-to-128 for M>128 (300->384); keep pow2 for M<=128. Byte-identical output (padding only). Baseline pp300/pp512/pp700/tg128: 30.66/40.42/40.17/2.58 t/s. Measured: 37.01/40.25/40.03/2.61 t/s. pp300 +20.7%; pp512/pp700 flat (ubatch<=512 caps M; 700=512+188, both bucketings map 188->256); tg unchanged.

2026-07-09 00:38:02 +02:00

..

cmake

ggml: backend-agnostic tensor parallelism (experimental) (#19378)

2026-04-09 16:42:19 +02:00

include

CUDA: lower-case PCI bus id, standardize for ggml (#22820)

2026-05-08 10:09:38 +02:00

src

rknpu2: M-bucketing — ceil-to-128 for M>128 instead of next_power_of_two

2026-07-09 00:38:02 +02:00

.gitignore

…

CMakeLists.txt

Merge tag 'b9101' into dspark

2026-07-08 21:00:48 +00:00