Search papers, labs, and topics across Lattice.
This paper introduces a novel approach to solving Bit Manipulation Puzzles in the NVIDIA Nemotron Model Reasoning Challenge by reframing logical deduction as a base-selection task that utilizes string similarity instead of complex arithmetic. By implementing a backtracking depth-first search (DFS) combined with autonomous error recovery, the authors effectively navigate the combinatorial explosion of possible operations while achieving over 96% validation accuracy. This method not only outperforms traditional techniques but also secures a 7th place finish in the competition, highlighting its effectiveness in logical reasoning tasks for LLMs.
Abandoning arithmetic logic for string similarity allows LLMs to achieve unprecedented accuracy in deducing logical rules from binary strings.
This paper presents our algorithmic innovations for the NVIDIA Nemotron Model Reasoning Challenge, focusing on Bit Manipulation Puzzles. In this task, the objective is to discover a hidden logical rule transforming input binary strings to outputs, then apply it to unseen inputs. Large Language Models (LLMs) notoriously struggle here; traditional methods force them to simulate complex boolean logic and arithmetic, leading to hallucinations. Furthermore, the search space of bitwise operations (combinations of shifts, rotations, and logic gates) suffers from a severe combinatorial explosion. To overcome this computational intractability, we present a novel approach that abandons arithmetic logic entirely in favor of string similarity, structured search, and autonomous error recovery. Our core contributions are: 1. Bases and Truth Table Formulation: We reframe logic-gate deduction into a base-selection task, leveraging string similarity (minimal bit flips) to isolate primitive transformations ("bases") and deduce truth tables without complex arithmetic. 2. Backtracking DFS and Error Recovery: We formalize a search process that tests candidate bases, detects logical collisions across examples, and backtracks upon failure to perform robust error recovery. 3. Bit Tokenization and Interactive Reasoning SFT: We force the tokenizer to encode binary strings as individual single-bit tokens. We use dynamic masking to simulate external oracle feedback, training the model to hypothesize, self-evaluate, and backtrack natively. Evaluated on bit manipulation puzzles, our approach achieved over 96% validation accuracy. This represents the highest performance in this category, driving our 7th Place overall finish in the contest.
NVIDIA