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

517ec60
Merge pull request #981 from JustVugg/fix/cuda-test-fmt7
Vincenzo Fornaro committed 3 months ago
bb310d4
fix(doctor): keep the Windows linkage probe portable in tests
Kenneth-Javier Ortigoza committed 3 months ago
36378ea
docs(plan): clarify Windows unified-memory discovery
Kenneth-Javier Ortigoza committed 3 months ago
dd6c22d
fix(cli): exclude unqualified GPUs from auto-enable
Kenneth-Javier Ortigoza committed 3 months ago
9ffa28b
fix(doctor): recognize the expected Windows GPU backend
Kenneth-Javier Ortigoza committed 3 months ago
ad8a75b
feat(plan): discover Windows AMD GPUs via hipInfo
Kenneth-Javier Ortigoza committed 3 months ago
80370ed
refactor(plan): isolate unqualified GPU discovery
Kenneth-Javier Ortigoza committed 3 months ago
b5a8bac
35a6037
fixup: <vector> for the fmt=7 smoke buffers
JustVugg committed 3 months ago
27f3f6c
185868f
Merge pull request #976 from JustVugg/fix/v4-ngen-clamp
Vincenzo Fornaro committed 3 months ago
8d1d96a
fadd9b5
360c57d
fix(build): compile CUDA backend on glibc >= 2.41
martinm86867-ops committed 3 months ago
863c4e6
7b50437
Merge pull request #943 from JustVugg/dev
Vincenzo Fornaro committed 3 months ago
72cd9f7
ci: run focused Python tests on Windows
Blake committed 3 months ago
1ba6f22
3780c4e
afad8b7