analiza aiqbench
This research investigates whether mechanistic interpretability signals can outperform standard heuristics like activation-norm or gradient-based methods for placing Parameter-Efficient Fine-Tuning (PEFT) budgets. The study finds that while causal selectors offer lower variance on content-sensitive tasks, simple magnitude-based approaches often match or exceed performance when output structure dominates, suggesting a nuanced diagnostic for practitioners rather than a universal replacement.
MechSparse: Mechanism-Guided Sparse PEFT Selection Is Task-Shaped
arXiv:2609.18961v1 Announce Type: new Abstract: Mechanistic interpretability identifies sparse subsets of heads and MLP blocks that carry specific behaviors. We ask whether such causal signals can guide where to place a small PEFT budget more effectively than the cheap heuristics practitioners already use. \method{} scores attention heads and MLP blocks by normalized activation-patching recovery on clean/corrupted probes and trains LoRA/QLoRA only on the selected sites; \methodc{} adds bounded credit for small within-layer joint subsets. We compare against random, magnitude, activation-norm, and gradient/Fisher on Ministral-8B/NF4 in three cells: Swahili span-JSON information extraction (IE) at $b{=}0.25\%$ and $1.0\%$, and English$\to$Swahili machine translation (MT) at $b{=}1.0\%$. The causal selectors never win the primary metric. On the headline IE cell (3 seeds, paired-bootstrap CIs over $600$ predictions), \methodc{} beats random by $+0.079$ span+type F1 and gradient/Fisher by $+0.174$, but trails activation-norm by $0.028$, with the smallest cross-seed std ($\pm 0.003$). On MT all four selectors lie within $0.30$ BLEU and every paired CI includes zero. A schema-versus-span decomposition explains the IE gap: activation-norm captures the rigid JSON routine, while causal scores track content-sensitive sites. We distill a preliminary diagnostic -- prefer activation-norm when output structure dominates, treat causal selectors as a hypothesis for content-dominated tasks -- and release masks, scores, predictions, and evaluation files for direct replay.
Abstract:Mechanistic interpretability identifies sparse subsets of heads and MLP blocks that carry specific behaviors. We ask whether such causal signals can guide where to place a small PEFT budget more effectively than the cheap heuristics practitioners already use. \method{} scores attention heads and MLP blocks by normalized activation-patching recovery on clean/corrupted probes and trains LoRA/QLoRA only on the selected sites; \methodc{} adds bounded credit for small within-layer joint subsets. We compare against random, magnitude, activation-norm, and gradient/Fisher on Ministral-8B/NF4 in three cells: Swahili span-JSON information extraction (IE) at $b{=}0.25\%$ and $1.0\%$, and English$\to$Swahili machine translation (MT) at $b{=}1.0\%$. The causal selectors never win the primary metric. On the headline IE cell (3 seeds, paired-bootstrap CIs over $600$ predictions), \methodc{} beats random by $+0.079$ span+type F1 and gradient/Fisher by $+0.174$, but trails activation-norm by $0.028$, with the smallest cross-seed std ($\pm 0.003$). On MT all four selectors lie within $0.30$ BLEU and every paired CI includes zero. A schema-versus-span decomposition explains the IE gap: activation-norm captures the rigid JSON routine, while causal scores track content-sensitive sites. We distill a preliminary diagnostic -- prefer activation-norm when output structure dominates, treat causal selectors as a hypothesis for content-dominated tasks -- and release masks, scores, predictions, and evaluation files for direct replay.
Computation and Language (cs.CL)
arXiv:2609.18961 [cs.CL]
(or arXiv:2609.18961v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.18961 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 · 303 cuvinte
Subiecte conexe
Links #5: 2026/07
LessWrong · 31 iul. 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · 13 oct. 2026
Performance Profiling on AMD GPUs – Part 5: Profiling-Driven Kernel Optimization with an AI Code-Assist Tool
AMD · 16 iul. 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · 8 oct. 2026