vllm
f3bdb0e0 - [Perf][MXFP4] Optimize activation quant for mxfp4. 3x kernel speedup, 5% throughput improvement (#63101)

Commit
24 days ago
[Perf][MXFP4] Optimize activation quant for mxfp4. 3x kernel speedup, 5% throughput improvement (#63101) Signed-off-by: ilmarkov <markovilya197@gmail.com> Co-authored-by: Roberto L. Castro <38211239+LopezCastroRoberto@users.noreply.github.com>
Author
Parents
Loading