AI Daybook
AI Daybook
오늘의 지면
논문연구·논문

Tree-Ising, 비트 수준 편집으로 언어 모델 수리

Tree-Ising: Repairing Language Models with Bit-Level Ising Edits

매체
Hugging Face
게시
중요도
2/5
화제 지수
21

한국어 요약

언어 모델의 가중치를 직접 읽고 수리하는 방식을 이산 최적화 문제로 변환하는 Tree-Ising 기법을 제시했다. BF16 가중치에 대한 부호 반전과 지수 시프트만으로 MATH-500에서 54.6%를 달성하며 GPT-6 Astra 에이전트 편집(51.7%)을 상회했다. 단 4번의 순전파 요청과 0개의 LLM 토큰만 소비했고, GPU 시간 약 0.90달러가 소요되었다.

원문 발췌 · English

Weight-space programming, introduced in recent MIT work by Takehi et al., enables a frontier LLM agent to read a network's weights and write a repair. Their GPT-6 Astra agent corrected the reasoning loops of Qwen3.5-0.8B without gradients, using 158 agent turns, 89.7 million tokens, and 50 forward-pass requests. Our method, Tree-Ising, casts this repair as a sparse discrete optimization problem over the model's own BF16 weights. Forward statistics define an objective that weakens a loop direction, strengthens a concluding direction, and preserves the layer's output on productive tokens. Within a state-tree representation, where lossless BF16 storage acts as a special case, the admissible edits are sign flips and exponent shifts that preserve all shared group exponents and mantissas. We decompose this objective exactly into independent QUBO (Ising) problems, one per output row, which we solve on a GPU using simulated annealing or quantum-inspired simulated bifurcation. On the agent's two layers, Tree-Ising achieves 54.6% on MATH-500, outperforming the agent's edit (51.7%) under the same evaluation protocol, while the base model scores 7.9% as reported. The simulated bifurcation solver yields a statistically indistinguishable repair at 55.1%. Both solvers outperform the agent edit across all five math benchmarks, with statistically significant improvements on MATH-500 and GSM8K after correcting for multiple comparisons. Our approach modifies 13% to 17% fewer weights and reduces BF16 bit flips by a factor of 2.2 to 2.3. It requires only four forward-pass requests and zero LLM tokens, with the annealing recipe costing about $0.90 in GPU time. On Qwen3.5-2B, with layers and dose selected using only the base model, the same recipe increases greedy GSM8K accuracy from 21.6% to 69.4%. As a proof of concept, we execute an eight-variable repair subproblem using QAOA on an IBM Heron processor.

#llm#optimization#model-repair#weight-editing
원문 기사 읽기
전체 보기 →