Quaff shows that activation outlier channels keep their spatial positions during LLM fine-tuning, and exploits this stability to cut fine-tuning memory and latency with INT8 quantization while matching or beating full-precision accuracy.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Quaff: Quantized Parameter-Efficient Fine-Tuning under Outlier Spatial Stability Hypothesis
Quaff shows that activation outlier channels keep their spatial positions during LLM fine-tuning, and exploits this stability to cut fine-tuning memory and latency with INT8 quantization while matching or beating full-precision accuracy.