The Post-Training Quantization Of Frantar
ArchiveMedium
Frantar et al. (2022) introduced a data-driven post-training quantization method using approximate second-order information, reducing GPT-family models to 4 bits with small quality drop. Name it (four-letter acronym). Flag format: CONGRESS{acronym}. Example: CONGRESS{awq}.
Show hint
The model family + 'quantization'.
Archive — no submissions accepted
This challenge is preserved for reference. Play live challenges at /challenges.