本文介绍了我们针对 NVIDIA Nemotron 模型推理挑战, 的算法创新，重点关注位操作难题。在此任务, 中，目标是发现隐藏的逻辑规则，将输入二进制字符串转换为输出,，然后将其应用于看不见的输入。大型语言模型 (LLMs) 在这方面表现不佳; 传统方法迫使它们模拟复杂的布尔逻辑和算术, 导致幻觉。此外，, 位运算的搜索空间( 移位, 旋转, 的组合和逻辑门) 遭受严重的组合爆炸。为了克服这种计算上的困难,，我们提出了一种新颖的方法，完全放弃算术逻辑，转而支持字符串相似性, 结构化搜索, 和自主错误恢复。我们的核心贡献是: 1. 碱基和真值表公式: 我们将逻辑门推导重新构建为碱基选择任务,，利用字符串相似性(最小位翻转)来隔离原始转换("碱基")并推导出真值表，无需复杂的算术。 2. 回溯 DFS 和错误恢复: 我们形式化了一个搜索过程，用于测试候选碱基, 检测示例, 之间的逻辑冲突，并在执行稳健错误恢复失败时进行回溯。 3. 位标记化和交互式推理 SFT: 我们强制标记器将二进制字符串编码为单独的单位标记。我们使用动态掩码来模拟外部预言机反馈, 训练模型来假设, 自评估, 并本地回溯。对位操作谜题进行评估,，我们的方法实现了超过 96% 的验证准确性。这代表了该类别中的最高性能。这些算法和训练创新最终为我们的团队赢得了, 竞赛中的最佳微调方法和最佳数据/合成数据方法奖，并在排行榜上获得了第七名。
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. These algorithmic and training innovations ultimately earned our team the Best Fine-tuning Method and Best Data/Synthetic Data Method awards in the contest, alongside a 7th Place overall finish on leaderboard.
科目:人工智能(cs.AI)
Subjects: Artificial Intelligence (cs.AI)