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
Merge pull request #1897 from JustVugg/docs/model-sizes
b86761b Vincenzo Fornaro committed 1 month ago
Merge remote-tracking branch 'origin/dev' into docs/model-sizes
9b92290 JustVugg committed 1 month ago
fix(serve): a past tool_calls that is not an array answered 500 on Qwen and Kimi
7c65c8d kevin9327 committed 1 month ago
Merge pull request #1896 from JustVugg/fix/setup-cuda-arch
a2ba7b1 Vincenzo Fornaro committed 1 month ago
docs: the README's welcome to the first discrete-GPU number, the credits kept, the setup's CUDA check
736082b JustVugg committed 1 month ago
docs: README rewritten around the one-step setup, in all five languages
11595fe JustVugg committed 1 month ago
setup: build CUDA only when the toolkit builds for the card (#1852)
8dae08a JustVugg committed 1 month ago
docs: the GLM-5.2 and Qwen3.6 download sizes as Hugging Face reports them
b14d9ba JustVugg committed 1 month ago
cluster: measure worker lease expiry with a monotonic clock
44f3176 Rudy Celekli committed 1 month ago
Merge pull request #1872 from JustVugg/feat/decide-clef
d63a60d Vincenzo Fornaro committed 1 month ago
fix(setup): confine downloader manifest writes to the model folder
5de9715 Rudy Celekli committed 1 month ago
cluster: report a disconnected peer without raising SIGPIPE
4e6b9f0 Rudy Celekli committed 1 month ago
test(download): reject HTTP error bodies with native curl
71c3535 Rudy Celekli committed 1 month ago