BitCal-TTS: Bit-Calibrated Test-Time Scaling for Quantized Reasoning Models
On the map: LLM reasoning → Test-time compute and efficient thinking
Abstract (arXiv)
Post-training quantization makes large reasoning models practical under tight memory and latency budgets, but it can distort the online signals that drive adaptive test-time compute allocation. Under a fixed cap on the number of newly generated tokens, miscalibrated confidence can lead to harmful early halting: the model may surface a plausible final line while the underlying reasoning is still wrong, or the controller may stop before the trace has stabilized. We study this interaction for greedy 4-bit inference and propose BitCal-TTS, a lightweight runtime controller that combines (i) inexpensive online proxies for token-level uncertainty and reasoning-trace stability, (ii) a bit-conditioned confidence rescaling that is conservative at low nominal precision, and (iii) a bit-aware post-marker confirmation horizon designed for GSM8K-style structured outputs. The method requires no fine-tuning of the base model and integrates with standard Hugging Face 4-bit inference using forward hooks for logits and last-layer hidden states. On small evaluation shards of GSM8K with Qwen2.5 Instruct models, BitCal-TTS improves exact-match accuracy over a non-bit-aware adaptive baseline at the 7B and 14B scales while preserving substantial token savings relative to fixed-budget decoding. At a token cap of B=512, on the evaluation shards we report (N=54 for 7B and N=35 for 14B; not the full GSM8K test set), accuracy gains are +3.7 points (7B) and +2.8 points (14B), with the premature-stop rate falling from 14.8% to 11.1% on 7B and from 17.1% to 11.4% on 14B. We report Wilson 95% confidence intervals throughout and explicitly discuss the limited statistical power of the partial-shard comparisons. We release code and figure-generation scripts to support full reproduction.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 10 of its 24 references that are in the map. First posted on arXiv 7 May 2026.
Builds on
10 of 24 references.
Closest ancestors · 4
Same direction, newest first
- 2025A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?31 Mar
- 2025s1: Simple test-time scaling31 Jan
- 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug
- 2022Language Models (Mostly) Know What They Know11 Jul
Other references · 6
- 2021Training Verifiers to Solve Math Word Problemscited by 10,109
- 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
- 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
- 2024Qwen2.5 Technical Reportcited by 4,942
- 2022LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scalecited by 906
- 2024A Careful Examination of Large Language Model Performance on Grade School Arithmeticcited by 191
Built on by
Citing papers per year: 2026: 1.
1 paper.
Similar, not cited
Close in wording, no citation either way
- 2026Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery
- 2026Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models
- 2026ReQAT: Achieving Full-Precision Reasoning Accuracy with 4-bit Floating-Point Quantization-Aware Training
- 2025Not All Bits Are Equal: Scale-Dependent Memory Optimization Strategies for Reasoning Models
- 2026What Makes Low-Bit Quantization-Aware Training Work for Reasoning LLMs? A Systematic Study
- 2026ReSET: Accurate Latency-Critical NVFP4 Reasoning via Step-Aware Temperature Scaling
- 2026Train Where the Quantized Model Goes: On-Policy Distillation for Low-Bit Reasoning
- 2025Quantization Hurts Reasoning? An Empirical Study on Quantized Reasoning Models