Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery

• URL: http://arxiv.org/abs/2606.02011v1
• Date: Mon, 01 Jun 2026 10:04:09 GMT
• Title: Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery
• Abstract summary: We show that aggressive 2-bit inference can fail to deliver end-to-end speedup because instability in the generation process inflates total token count.<n>We analyze full reasoning traces of Qwen3 reasoning models across mathematical and commonsense benchmarks.<n>Our results show that extreme low-bit reasoning becomes practical when its failures are treated as controllable generation pathologies.
• Score: 40.94400211806987
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Reasoning Models (LRMs) rely on long reasoning traces, making inference expensive. While low-bit quantization reduces per-token decoding cost, we show that aggressive 2-bit inference can fail to deliver end-to-end speedup because instability in the generation process inflates total token count. Instead of merely lowering answer accuracy, 2-bit quantization often produces much longer traces with repetitive loops, budget exhaustion, delayed commitment, and unclosed reasoning segments. We analyze full reasoning traces of Qwen3 reasoning models across mathematical and commonsense benchmarks and show that accuracy degradation is tightly linked to these process-level failures. To address them, we introduce two lightweight controls: FP16 planning, which gives the 2-bit model a short high-precision outline, and loop rescue, which detects repetitive traces and either commits to an earlier answer or falls back to FP16. On MATH-500, loop rescue improves Qwen3-8B accuracy from 17.2% to 74.2%, while planning plus loop rescue improves Qwen3-32B from 65.0% to 87.2%. Overall, our results show that extreme low-bit reasoning becomes practical when its failures are treated as controllable generation pathologies: with lightweight detection and selective FP16 support, 2-bit inference can recover accuracy while preserving real end-to-end speed. Our code is available at: https://github.com/brain-lab-research/quantized-reasoning.

Related papers

• Wrong Prediction, Right Answer: Recovering Evidence from Collapsed LLM Sequence Scores [31.709549159768727]
  We observe a consistent readout gap across diverse reasoning benchmarks.<n> hidden-state probes successfully decode correct answers even when native sequence scoring completely collapses.<n>Our results show that many apparent zero-shot reasoning deficits are expression failures masking intact internal logic.
  arXiv  Detail & Related papers  (2026-08-31T16:43:55Z)
• STOP: Structured On-Policy Pruning of Long-Form Reasoning in Low-Data Regimes [13.293115227628775]
  Long chain-of-thought (Long CoT) reasoning improves performance on multi-step problems, but it also induces overthinking.<n>We propose STOP (Structured On-policy Pruning), an on-policy algorithm for analyzing and pruning long-form reasoning traces.
  arXiv  Detail & Related papers  (2026-05-13T08:28:05Z)
• Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning [11.179446105672461]
  We propose a multi-stage efficient reasoning method that combines supervised fine-tuning and reinforcement learning.<n>Our approach reduces response length by an average of 28% for 8B models and 40% for 32B models.<n>It achieves a superior trade-off compared to more complex state-of-the-art efficient reasoning methods.
  arXiv  Detail & Related papers  (2026-01-06T12:31:51Z)
• DTS: Enhancing Large Reasoning Models via Decoding Tree Sketching [54.98126916293868]
  Large Reasoning Models (LRMs) produce excessively long chain-of-thought traces that degrade accuracy.<n>We propose a model-agnostic decoding framework that sketches the reasoning space by branching at high-entropy tokens and applies early stopping to select the shortest completed reasoning path.<n>This approach approximates the optimal solution that enhances both efficiency and accuracy, without requiring additional training or supervision.
  arXiv  Detail & Related papers  (2025-11-01T17:41:28Z)
• DLER: Doing Length pEnalty Right - Incentivizing More Intelligence per Token via Reinforcement Learning [134.03095505580276]
  Doing Length pEnalty Right (DLER) is a training recipe combining batch-wise reward normalization, higher clipping, dynamic sampling, and a simple truncation length penalty.<n>DLER achieves state-of-the-art accuracy--efficiency trade-offs, cutting output length by over 70 percent while surpassing all previous baseline accuracy.
  arXiv  Detail & Related papers  (2025-10-16T20:05:57Z)
• Abduct, Act, Predict: Scaffolding Causal Inference for Automated Failure Attribution in Multi-Agent Systems [20.846301581161978]
  Failure attribution in multi-agent systems is a critical yet unsolved challenge.<n>Current methods treat this as a pattern recognition task over long conversation logs.<n>A2P Scaffolding transforms failure attribution from pattern recognition into a structured causal inference task.
  arXiv  Detail & Related papers  (2025-09-12T16:51:15Z)
• Fast Quiet-STaR: Thinking Without Thought Tokens [51.79231070632772]
  Fast Quiet STaR is a more efficient reasoning framework that preserves the benefits of token-level reasoning while reducing computational cost.<n>Our method introduces a curriculum learning based training strategy that gradually reduces the number of thought tokens.<n>Experiments on four benchmark datasets with Mistral 7B and Qwen2.5 7B demonstrate that Fast Quiet-STaR consistently outperforms Quiet-STaR in terms of average accuracy.
  arXiv  Detail & Related papers  (2025-05-23T11:14:12Z)
• Pushing the Limits of Low-Bit Optimizers: A Focus on EMA Dynamics [64.62231094774211]
  Statefuls (e.g., Adam) maintain auxiliary information even 2x the model size in order to achieve optimal convergence.<n>SOLO enables Adam-styles to maintain quantized states with precision as low as 3 bits, or even 2 bits.<n>SOLO can thus be seamlessly applied to Adam-styles, leading to substantial memory savings with minimal accuracy loss.
  arXiv  Detail & Related papers  (2025-05-01T06:47:45Z)
• O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning [98.3430004984531]
  We propose Length-Harmonizing Fine-Tuning (O1-Pruner) to minimize reasoning overhead while maintaining accuracy.<n>Our code is coming soon at https://github.com/StarDewXXX/O1-Pruner.
  arXiv  Detail & Related papers  (2025-01-22T01:35:11Z)
• QERA: an Analytical Framework for Quantization Error Reconstruction [12.110441045050223]
  An increasing interest in quantizing weights to extremely low precision while offsetting the resulting error with low-rank, high-precision error reconstruction terms.<n>The combination of quantization and low-rank approximation is now popular in both adapter-based, parameter-efficient fine-tuning methods.<n>We formulate an analytical framework, named Quantization Error Reconstruction Analysis (QERA), and offer a closed-form solution to the problem.
  arXiv  Detail & Related papers  (2024-10-08T13:37:34Z)

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.