Reasoning-Trace Collapse: Evaluating the Loss of Explicit Reasoning During Fine-Tuning
For practitioners fine-tuning reasoning models, this work highlights a critical evaluation gap and offers a simple mitigation, though the problem is domain-specific to explicit reasoning models.
Fine-tuning reasoning models on ordinary instruction-response data can cause reasoning-trace collapse, where models lose structurally valid reasoning traces while maintaining plausible final answers. The study shows that answer-only metrics obscure this failure, and loss-masking strategies can mitigate collapse.
Explicit reasoning models are trained to produce intermediate reasoning traces before final answers, but downstream fine-tuning is often performed on ordinary instruction-response data that contains no such traces. We show that this mismatch can induce reasoning-trace collapse: a fine-tuned model continues to produce plausible final answers while losing the structurally valid explicit reasoning traces that made it a reasoning model in the first place. We introduce a structural evaluation framework that separates answer correctness from reasoning-trace validity, measuring valid, empty, missing, and truncated reasoning alongside reasoning-conditioned task performance. Using this framework, we study four open-weight reasoning models and find that standard supervised fine-tuning can rapidly suppress valid reasoning traces, and that answer-only metrics can substantially obscure this failure: in several settings, performance conditional on valid reasoning remains high while the rate of valid reasoning falls sharply. We further show that simple loss-masking strategies can substantially mitigate collapse without requiring teacher-generated reasoning traces. These results suggest that evaluations of fine-tuned reasoning models should report structural reasoning reliability metrics in addition to final-answer performance, especially when adaptation data does not contain explicit reasoning traces.