RILQ improves 2-bit LLM accuracy by optimizing a model-wise discrepancy loss at the final layer, making LoRA error compensation rank-insensitive.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
RILQ: Rank-Insensitive LoRA-based Quantization Error Compensation for Boosting 2-bit Large Language Model Accuracy
RILQ improves 2-bit LLM accuracy by optimizing a model-wise discrepancy loss at the final layer, making LoRA error compensation rank-insensitive.