An 8B LLM fine-tuned on certifiably feasible/infeasible minor-embedding data jointly generates solutions and detects infeasibility, yielding up to 30% higher accuracy than GPT-5.2 and 2x speedup in downstream local search.
Objective Among all feasible valid embeddings, minimize the total number of hardware nodes utilized across all chains (i.e., minimize the sum of|S i|for alli)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Infeasibility Aware Large Language Models for Combinatorial Optimization
An 8B LLM fine-tuned on certifiably feasible/infeasible minor-embedding data jointly generates solutions and detects infeasibility, yielding up to 30% higher accuracy than GPT-5.2 and 2x speedup in downstream local search.