Search papers, labs, and topics across Lattice.
This paper investigates the generalization discrepancy observed when fine-tuning large language models (LLMs) on Chain-of-Thought (CoT) data generated by different source models, DeepSeek-R1-0528 and gpt-oss-120b. They found that despite achieving lower training loss, models fine-tuned on DeepSeek-R1-0528's CoT data generalize worse due to its divergent reasoning patterns compared to gpt-oss-120b's convergent approach. By filtering out frequently branching trajectories from the DeepSeek-R1-0528 dataset, the authors demonstrate improved generalization performance on reasoning benchmarks.
LLMs can learn to reason *worse* from seemingly better training data: models trained on CoT data with lower loss can generalize poorly due to inheriting inefficient, divergent reasoning patterns.
Supervised Fine-Tuning (SFT) on long Chain-of-Thought (CoT) trajectories has become a pivotal phase in building large reasoning models. However, how CoT trajectories from different sources influence the generalization performance of models remains an open question. In this paper, we conduct a comparative study using two sources of verified CoT trajectories generated by two competing models, \texttt{DeepSeek-R1-0528} and \texttt{gpt-oss-120b}, with their problem sets controlled to be identical. Despite their comparable performance, we uncover a striking paradox: lower training loss does not translate to better generalization. SFT on \texttt{DeepSeek-R1-0528} data achieves remarkably lower training loss, yet exhibits significantly worse generalization performance on reasoning benchmarks compared to those trained on \texttt{gpt-oss-120b}. To understand this paradox, we perform a multi-faceted analysis probing token-level SFT loss and step-level reasoning behaviors. Our analysis reveals a difference in reasoning patterns. \texttt{gpt-oss-120b} exhibits highly convergent and deductive trajectories, whereas \texttt{DeepSeek-R1-0528} favors a divergent and branch-heavy exploration pattern. Consequently, models trained with \texttt{DeepSeek-R1} data inherit inefficient exploration behaviors, often getting trapped in redundant exploratory branches that hinder them from reaching correct solutions. Building upon this insight, we propose a simple yet effective remedy of filtering out frequently branching trajectories to improve the generalization of SFT. Experiments show that training on selected \texttt{DeepSeek-R1-0528} subsets surprisingly improves reasoning performance by up to 5.1% on AIME25, 5.5% on BeyondAIME, and on average 3.6% on five benchmarks.
