Skip to content

Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

C

Commits on main

abe3ca8
fix(qwen36): keep int8 expert weights the tier still points at (#1341)
Courtney Andrew Richardson committed 2 months ago
d937ad2
fix(qwen36): wake cv_take waiters on tier shutdown (#1340)
Courtney Andrew Richardson committed 2 months ago
246ce2c
fix(qwen36): size the tier's input replicas per device (#1339)
Courtney Andrew Richardson committed 2 months ago
b5f6f1a
f58a267
9b13d3a
25ecb13
d9e0a8c
fix(doctor): omit GPU plan for CPU-only engine
Mauricio Sevilla Britto committed 2 months ago
700151d
cfd5106
f01621f
docs: explain hybrid CUDA OpenMP override (#1293)
ZacharyZcR committed 2 months ago
e8e4cc0
docs: clarify rANS record extent (#1292)
ZacharyZcR committed 2 months ago
b504c68
a642a8b
Merge pull request #1303 from JustVugg/dev
Vincenzo Fornaro committed 2 months ago
12a5c46