analiza aiqbench
QK-Wanda introduces a novel pruning strategy for large language models that couples query and key weight scoring to share a budget, achieving significant reductions in reconstruction error without requiring gradient updates. While the method offers faster inference potential through aggressive sparsity, results indicate that lower local reconstruction error does not always guarantee improved downstream performance across all model architectures. This research highlights the nuanced relationship between structural optimization metrics and actual task accuracy in model compression efforts.
QK-Wanda: Coupling Queries and Keys for Unstructured Pruning
arXiv:2610.01554v1 Announce Type: cross Abstract: Wanda (Sun et al., 2024) prunes large language models by scoring weights independently within each linear projection, although queries and keys interact through dot products. We introduce QK-Wanda, which scores query and key weights by their individual deletion costs under an unmasked pre-RoPE reconstruction objective. It augments Wanda scores with information from the opposite projection (keys for query weights, and queries for key weights), allowing both projections to share a pruning budget. Its closed-form scores require no gradients or weight updates; full pruning takes 1.3% longer than Wanda on A100 and 3.1% longer on H200 with the calibration used in our main experiments. We evaluate QK-only pruning across 15 models from TinyLlama, Llama 2, Llama 3, and Qwen2.5, spanning 0.5B-72B parameters. Relative to Wanda, QK-Wanda reduces QK reconstruction error by an average of 60% at 50% sparsity and 45% at 80%. Downstream gains depend on the model. At 80% sparsity on Llama 2 70B, WikiText-2 and C4 perplexity decrease by 20.3% and 13.5%, while mean zero-shot accuracy rises by 5.94 percentage points. Qwen2.5-72B also improves, but Llama-3.1-70B has substantially higher perplexity despite lower reconstruction error. These results show both the promise of coupled pruning criteria and the limits of local reconstruction as a predictor of model quality.
Abstract:Wanda (Sun et al., 2024) prunes large language models by scoring weights independently within each linear projection, although queries and keys interact through dot products. We introduce QK-Wanda, which scores query and key weights by their individual deletion costs under an unmasked pre-RoPE reconstruction objective. It augments Wanda scores with information from the opposite projection (keys for query weights, and queries for key weights), allowing both projections to share a pruning budget. Its closed-form scores require no gradients or weight updates; full pruning takes 1.3% longer than Wanda on A100 and 3.1% longer on H200 with the calibration used in our main experiments. We evaluate QK-only pruning across 15 models from TinyLlama, Llama 2, Llama 3, and Qwen2.5, spanning 0.5B-72B parameters. Relative to Wanda, QK-Wanda reduces QK reconstruction error by an average of 60% at 50% sparsity and 45% at 80%. Downstream gains depend on the model. At 80% sparsity on Llama 2 70B, WikiText-2 and C4 perplexity decrease by 20.3% and 13.5%, while mean zero-shot accuracy rises by 5.94 percentage points. Qwen2.5-72B also improves, but Llama-3.1-70B has substantially higher perplexity despite lower reconstruction error. These results show both the promise of coupled pruning criteria and the limits of local reconstruction as a predictor of model quality.
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2610.01554 [cs.LG]
(or arXiv:2610.01554v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.01554 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 305 cuvinte
Subiecte conexe
PySpark Optimization: 12 Proven Techniques to Speed Up Your Spark Jobs
Analytics Vidhya · 27 mai 2026
Compositional Chain-of-Relations for Faithful Knowledge Graph Question Answering with Large Language Models
arXiv · 25 aug. 2026
GraphRAG with TypeSafe Jev: A System One Approach to Scalable Knowledge Graphs
Towards Data Science · 27 sept. 2026
Magnitude Profile Pruning: Calibration-Free Structured Attention Head Removal for Transformer Compression
arXiv · 24 sept. 2026