Categories / LLM Infrastructure

SSD-streaming inference engine for giant MoE models (Rust + CUDA). GLM 5.2 743B at 2 tok/s and Hy3 295B at 7 tok/s on two consumer 16GB GPUs. Zero-config multi-GPU: measures PCIe bandwidth, places attention and hot experts where they fit.

Last push 2 months ago

cudaggufglminference-enginellmlocal-llmmixture-of-expertsmoemulti-gpuquantizationrustssd-streaming

Created

Jul 2026

Total stars

239

Last 7 days

0

Last 30 days

0

Last 12 months

0

README badge

Rusty Stars badge for giannisanni/pulsar
[![Rusty Stars](https://rustystars.dev/badge/giannisanni/pulsar.svg)](https://rustystars.dev/giannisanni/pulsar)