analiza aiqbench
F-WANDA presents a significant advancement in sustainable LLM deployment by optimizing post-training pruning through Fisher information, effectively balancing model quality and computational cost without requiring weight updates or extra calibration data. This method achieves superior performance on standard benchmarks while drastically reducing the energy and time overhead associated with traditional sparsification techniques like SPARSEGPT. By placing this approach on the Pareto frontier of compression efficiency, it offers a practical pathway for deploying large language models in resource-constrained environments.
F-WANDA: Fisher-Reweighted Post-Training Pruning for Sustainable Deployment of Large Language Models
arXiv:2608.00481v1 Announce Type: new Abstract: One-shot post-training pruning is the most energy-frugal compression strategy for largelanguage models (LLMs), yet existing approaches trade either quality (WANDA) or compute cost (SPARSEGPT). We introduce F-WANDA, a drop-in modification of WANDA that reallocates the per-row keep budget across output neurons in proportion to the empirical Fisher information of the pre-activation. The Fisher signal is collected in a single additional backward pass over the same calibration corpus WANDA already uses; no weights are updated. On LLAMA-2-7B at 50 % unstructured sparsity, F-WANDA attains WikiText-2 perplexity of 6.85, matches WANDA fluency, and improves 5-shot MMLU by +1.6 pp over WANDA and +1.1 pp over SPARSEGPT, while incurring only one-third of SPARSEGPT pruning wall-clock and energy. The headline trade-off is achieved without extra calibration data or fine-tuning, placing F-WANDA on the Pareto frontier of quality versus pruning cost for sustainable LLM compression.
Abstract:One-shot post-training pruning is the most energy-frugal compression strategy for largelanguage models (LLMs), yet existing approaches trade either quality (WANDA) or compute cost (SPARSEGPT). We introduce F-WANDA, a drop-in modification of WANDA that reallocates the per-row keep budget across output neurons in proportion to the empirical Fisher information of the pre-activation. The Fisher signal is collected in a single additional backward pass over the same calibration corpus WANDA already uses; no weights are updated. On LLAMA-2-7B at 50 % unstructured sparsity, F-WANDA attains WikiText-2 perplexity of 6.85, matches WANDA fluency, and improves 5-shot MMLU by +1.6 pp over WANDA and +1.1 pp over SPARSEGPT, while incurring only one-third of SPARSEGPT pruning wall-clock and energy. The headline trade-off is achieved without extra calibration data or fine-tuning, placing F-WANDA on the Pareto frontier of quality versus pruning cost for sustainable LLM compression.
Artificial Intelligence (cs.AI)
arXiv:2608.00481 [cs.AI]
(or arXiv:2608.00481v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2608.00481 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 · 234 cuvinte
Subiecte conexe
Quantization and Pruning Methods to Make Your LLM Leaner
KDnuggets · 28 aug. 2026
How to Size GPUs for AI Inference and TCO Without Overspending
NVIDIA · 1 sept. 2026
Dynamic Entanglement-Weighted Pruning for Quantum Federated Unlearning in Supply-Chain Risk Prediction
arXiv · 20 aug. 2026
Links #3: 2026/06 Part 1
LessWrong · 15 iun. 2026