Reasoning-Trace Collapse: Evaluating the Loss of Explicit Reasoning During Fine-Tuning

• URL: http://arxiv.org/abs/2605.21127v1
• Date: Wed, 20 May 2026 12:58:01 GMT
• Title: Reasoning-Trace Collapse: Evaluating the Loss of Explicit Reasoning During Fine-Tuning
• Abstract summary: We show that a fine-tuned model continues to produce plausible final answers while losing the structurally valid explicit reasoning traces.<n>We introduce a structural evaluation framework that separates answer correctness from reasoning-trace validity.<n>We show that simple loss-masking strategies can substantially collapse without requiring teacher-generated reasoning traces.
• Score: 9.426248468755272
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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.

Related papers

• DERELAB: Probing Defeasible Reasoning and Confirmation Bias in LLMs with a Generative Benchmark [6.835919520017402]
  Defeasible reasoning is a type of reasoning where inferences are drawn from plausible current evidence, but can be retracted upon the introduction of newer evidence.<n>We introduce DeReLab, a generative framework that produces multi-turn belief-updating conversations from parameterized graph structures.<n>We evaluate nine open and proprietary large language models and find that nearly all exhibit a systematic tendency to accept congruent evidence while resisting incongruent updates.
  arXiv  Detail & Related papers  (2026-08-31T08:10:29Z)
• Correct Is Not Enough: Training Reasoning Planners with Executor-Grounded Rewards [15.458797431655961]
  Reinforcement learning with verifiable rewards has become a common way to improve explicit reasoning in large language models.<n>But final-answer correctness alone does not reveal whether the reasoning trace is faithful, reliable, or useful to the model that consumes it.<n>We propose TraceLift, a planner-executor training framework that treats reasoning as a consumable intermediate artifact.
  arXiv  Detail & Related papers  (2026-05-05T15:24:40Z)
• Shorthand for Thought: Compressing LLM Reasoning via Entropy-Guided Supertokens [4.0704009036918025]
  Reasoning in Large Language Models incurs significant inference-time compute.<n>We show that reasoning tokens split into two functional types: low-entropy textitstructural tokens (recurring phrases that scaffold the reasoning process) and higher-entropy textitorganic tokens (problem-specific content that drives toward a solution)<n>We derive textitsupertokens that capture frequent structural patterns, then teach the model to adopt them via supervised fine-tuning.
  arXiv  Detail & Related papers  (2026-04-29T07:06:43Z)
• Early Stopping for Large Reasoning Models via Confidence Dynamics [55.67938134245981]
  Large reasoning models rely on long chain-of-thought generation to solve complex problems.<n>Key challenge is determining when the model should stop reasoning and produce the final answer.<n>We propose CoDE-Stop, an early stopping method that leverages the dynamics of intermediate answer confidence to decide when to terminate reasoning.
  arXiv  Detail & Related papers  (2026-04-06T17:59:45Z)
• CausalFlip: A Benchmark for LLM Causal Judgment Beyond Semantic Matching [50.65932158912512]
  We propose a new causal reasoning benchmark, CausalFlip, to encourage the development of new large language models.<n>CaulFlip consists of causal judgment questions built over event triples that could form different confounder, chain, and collider relations.<n>We evaluate LLMs under multiple training paradigms, including answer-only training, explicit Chain-of-Thought supervision, and a proposed internalized causal reasoning approach.
  arXiv  Detail & Related papers  (2026-02-23T18:06:15Z)
• Structured Reasoning for Large Language Models [59.215789462977206]
  We propose Structured Reasoning (SCR), a framework that decouples reasoning trajectories into explicit, evaluable, and trainable components.<n>SCR substantially improves reasoning efficiency and self-verification.<n>Compared with existing reasoning paradigms, it reduces output token length by up to 50%.
  arXiv  Detail & Related papers  (2026-01-12T04:04:01Z)
• Do LLMs Encode Functional Importance of Reasoning Tokens? [11.21558453188654]
  We propose greedy pruning, a likelihood-preserving deletion procedure that iteratively removes reasoning tokens.<n>We show that students trained on pruned chains outperform a frontier-model-supervised compression baseline at matched reasoning lengths.
  arXiv  Detail & Related papers  (2026-01-06T14:50:02Z)
• Stop Spinning Wheels: Mitigating LLM Overthinking via Mining Patterns for Early Reasoning Exit [114.83867400179354]
  Overthinking can degrade overall performance of large language models.<n>We categorize reasoning into three stages: insufficient exploration stage, compensatory reasoning stage, and reasoning convergence stage.<n>We develop a lightweight thresholding strategy based on rules to improve reasoning accuracy.
  arXiv  Detail & Related papers  (2025-08-25T03:17:17Z)
• Think Clearly: Improving Reasoning via Redundant Token Pruning [57.01254508252785]
  We show that deliberately removing redundancy in the reasoning process significantly improves performance.<n>We demonstrate that our method significantly improves overall accuracy across reasoning-intensive benchmarks without any training.
  arXiv  Detail & Related papers  (2025-06-17T06:04:01Z)
• When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMs [19.354141845315276]
  Chain-of-thought reasoning can significantly degrade instruction-following accuracy.<n>This is the first work to systematically expose reasoning-induced failures in instruction-following.
  arXiv  Detail & Related papers  (2025-05-16T16:36:00Z)
• MetaLogic: Logical Reasoning Explanations with Fine-Grained Structure [129.8481568648651]
  We propose a benchmark to investigate models' logical reasoning capabilities in complex real-life scenarios. Based on the multi-hop chain of reasoning, the explanation form includes three main components. We evaluate the current best models' performance on this new explanation form.
  arXiv  Detail & Related papers  (2022-10-22T16:01:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.