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

inkling: chunked pread for >2GB resident tensors
Nicholas Beerbower committed 4 months ago
32b97a7
convert_inkling: completion check counts only model-* shards
Nicholas Beerbower committed 4 months ago
de12586
593bc3d
inkling: bf16-resident weights + AVX-512 BF16/VNNI kernels
Nicholas Beerbower committed 4 months ago
92bdb87
tools: Inkling bf16 -> colibri int4 converter, e2e-validated
Nicholas Beerbower committed 4 months ago
b721a1e
Merge pull request #534 from JustVugg/dev
Vincenzo Fornaro committed 4 months ago
81f08a0
Merge pull request #533 from JustVugg/release/v1.1.1
Vincenzo Fornaro committed 4 months ago
cd278c2
Merge pull request #532 from JustVugg/fix/grdraft-bss-blob
Vincenzo Fornaro committed 4 months ago
6164f9f
Merge pull request #530 from JustVugg/fix/release-sha256sums
Vincenzo Fornaro committed 4 months ago
7f77d39
16aeb53
feat: add learned partial mirror planner
Bill Herald committed 4 months ago
0324d10
9baae9b
a04721e
83522db
docs(v4): track dynamic DSpark tuning
DrewZ committed 4 months ago
db1463e
docs: add Chinese DeepSeek V4 guide
DrewZ committed 4 months ago
0c17d02
29f7a3c