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).
About
from-scratch LLM inference for RTX 5090 (sm_120a) and H100 (sm_90a)
Languages
Contributors3
![imgbot[bot]](https://avatars.githubusercontent.com/in/4706?v=4)
No features listed.
Comments Theme
