3 ms·

GPT-OSS-120B runs like hell on my DGX Spark

• embedding-shape 8mo ago
  The MXFP4 variant I suppose? My setup (RTX Pro 6000) does around ~140 tok/s with llama.cpp, around 160 tok/s with vLLM.
  • vkaufmann 8mo ago
    yep MXFP4 really fast :D