Search results

Per-layer KV heads, attention logit capping, MoE per-expert scales, NPU multi-input

Status: Open (in progress).

cactus-compute/cactusnumber 526

#526 In cactus-compute/cactus;

· ncylich opened on Mar 19, 2026

reverting attn exp calculations to before 3n

Status: Open (in progress).

cactus-compute/cactusnumber 511

#511 In cactus-compute/cactus;

· ncylich opened on Mar 9, 2026

Optimized Attention

Status: Open (in progress).

cactus-compute/cactusnumber 480

#480 In cactus-compute/cactus;

· ncylich opened on Mar 2, 2026

Benchmarking against other quantized kernels

Status: Open (in progress).

cactus-compute/cactusnumber 458

#458 In cactus-compute/cactus;

· ncylich opened on Feb 26, 2026