bw24
by avifenesh · github.com/avifenesh/bw24
From-scratch Rust+CUDA inference engine, bit-exact by construction — NVFP4, MoE, MTP speculative decoding, tuned against measured limits of one RTX 5090 Laptop (sm_120a).
Stars
★ 281
Forks
35
Language
Rust
Category
Dev Tools
License
MIT
blackwellcudaggufgpu-kernelsllama-cppllm-inferencemoenvfp4rustspeculative-decoding
Featured in
- bw24 @ 10:08 · Jul 9, 2026