Archive
Fine-Tuning & Training

The Quantization That Zero-Crashes

Archive
Hard
200pts0 solves
Lin et al. (2023) proposed a weight-only post-training quantization method that preserves the salient 1% of weights by activation-aware scaling, used in llama.cpp and vLLM. Three-letter acronym?
Show hint
Activation + 'weight' + 'quantization', initials.

Archive — no submissions accepted

This challenge is preserved for reference. Play live challenges at /challenges.