vllm
f3bdb0e0
- [Perf][MXFP4] Optimize activation quant for mxfp4. 3x kernel speedup, 5% throughput improvement (#63101)
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
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>
References
#63101 - [Perf][MXFP4] Optimize activation quant for mxfp4. 3x kernel speedup, 5% throughput improvement
Author
ilmarkov
Parents
01b05cd6
Loading