IST-DASLab/marlin: FP16xINT4 LLM inference kernel
repo
Fields
id
sources/quant-marlin-repoPublisher
IST-DASLab
Kind
repo
Published
2024-01-17
Accessed
2026-10-03
notes
Mixed-precision 4-bit weight-only kernel; upstream README claims ~4x speedup up to batch 16-32.
Notes
Mixed-precision 4-bit weight-only kernel; upstream README claims ~4x speedup up to batch 16-32.
Cross-references
This record references no other records.