Search

17 results for “MATH-500”

1. Research · Efficiency & systems

   Scalar two-bit weight quantization can preserve language model reasoning accuracy

   ThinkBit: Search Codewords, Recalibrate Scales, Select by Reasoning

   ThinkBit, a five-stage framework, keeps MATH-500 accuracy within a few points of bfloat16 on models up to 14B while shrinking memory traffic by up to 8x.

   MATH-500: 46.7% accuracy (as reported)

   27 Oct 2026Microsoft Applied Sciences Group

2. Research · Language models

   Starting token cues can elicit reasoning from base language models

   Base Models Can Reason By Taking a Cue From Training Data

   Fixing just the first two tokens of a response dramatically improves performance on math and coding, rivaling reinforcement learning.

   MATH-500: 16.5% pass@1 (as reported)

   6 Oct 2026

3. Research · Language models

   LLM pipelines can lose 40.5 accuracy points between stages

   The Decomposition Tax: LLM Pipelines Lose Up to 40 Accuracy Points at Their Own Interfaces

   Across mathematical reasoning benchmarks, preserving access to the original problem substantially improved multi-stage pipelines without changing the model, prompts, stages or completion budget.

   MATH-500: 55.5% accuracy (as reported)

   29 Sept 2026Rutgers University, Nanjing Tech University, Nanjing University of Posts and Telecommunications

4. Research · Language models

   Training on self-generated answers improves reasoning in low-bit language models

   Right On Q: Recovering Reasoning under Quantization via Self-Generated Trajectories

   Across three Qwen3 model sizes at 2- and 3-bit precision, OnQ improves benchmark accuracy while reducing truncated and repetitive responses.

   MATH-500: 69.33% accuracy (as reported)

   22 Oct 2026University of Michigan, Ann Arbor, University of California, Santa Barbara, Meta AI

5. Research · Language models

   Sparse RL-Guided Token Choices Recover Most Math Reasoning Gains

   What RL Brings to Reasoning: A Few Decisions the Base Model Would Not Make

   Replacing about 4% of a base model’s generated tokens with an RL-trained model’s choices recovered 90.5% of the reinforcement-learning improvement on AIME.

   AIME 2022-2025: 19.32% pass@1 (as reported)

   20 Oct 2026University of California, San Diego, Meituan, LongCat Team, The Chinese University of Hong Kong, Shenzhen

6. Research · Language models

   Cross-feedback and step-wise scoring let small models match much larger reasoning datasets

   Collaborative Reasoning Distillation via Cross-Feedback and Coherent Curation

   The authors' CRD framework iteratively critiques multiple teachers, scores individual reasoning steps, and stitches complementary chains to train compact models on 50K examples.

   MATH-500: 97.3% pass@1 (as reported)

   9 Oct 2026KAIST

7. Research · Language models

   Continuous diffusion language models match discrete peers while enabling smoother control

   Large Language Continuous Diffusion Models

   Sigma scales continuous language diffusion to 3B and 8B parameters, using 16-dimensional token embeddings to support controllable sampling and few-step generation.

   MATH-500: 70.44% accuracy (as reported)

   5 Oct 2026NVIDIA, Cornell University, Georgia Institute of Technology

8. Research · Efficiency & systems

   Low-rank KV cache compression preserves all tokens during long reasoning

   iS-KV: Online Low-Rank KV Cache Compression via Block-Incremental SVD

   iS-KV uses block-incremental SVD to update both basis and coefficients, avoiding representation drift and matching original model accuracy at 4-5x compression on DeepSeek-R1 and Qwen3-8B.

   MATH-500: 82.6% accuracy (as reported)

   5 Oct 2026The Hong Kong University of Science and Technology (Guangzhou), Guangdong OPPO Mobile Telecommunications Corp., Ltd., Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences

9. Research · Agents

   Stepwise LLM routing cuts multi-agent workflow costs while preserving quality

   ProgRouter: Online Progress-Guided Orchestration for Multi-Agent LLM Workflows under Quality-Cost Tradeoffs

   ProgRouter selects an LLM at each workflow step using estimated task progress, time constraints, and long-term cost efficiency.

   27 Aug 2026Aston University, Queen Mary University of London, University of Exeter

10. Research · Efficiency & systems

    Language models can adapt compressed memory without growing their dictionaries

    FAD-KV: Adaptive Dictionary KV Caching for Efficient Reasoning Decoding

    The authors combine faster selection of compression components with reuse of inactive components, keeping the dictionary fixed in size as generation continues.

    MATH-500: 85.7% full-OMP support recall (as reported)

    30 Oct 2026Rensselaer Polytechnic Institute, IBM Research

11. Research · Language models

    Language models gain from varied answers, not necessarily altered weights

    Rethinking Neural Thickets Without Weight Perturbations

    Dufour and colleagues show that changing output scores, selecting useful variants and combining their answers can match or exceed ensembles built by perturbing model weights.

    MATH-500: 63.3% accuracy (as reported)

    30 Oct 2026Kyutai, Meta FAIR, LIGM

12. Research · Language models

    Reasoning models usually inherit “wait” from training data, not reinforcement learning

    Inherited, Not Invented: Tracing the Provenance of Wait in Reasoning Models

    Across open-model training histories, the authors find that reinforcement learning can improve mathematics accuracy without introducing “wait”, while different training recipes amplify or erase an inherited marker.

    MATH-500: 41% accuracy (as reported)

    23 Oct 2026The Hong Kong University of Science and Technology, Huawei Noahah’s Ark Lab, Peking University

13. Research · Language models

    Uncertainty-weighted credit improves cross-tokenizer distillation for LLMs

    CAST: Credit Alignment for Cross-Tokenizer On-Policy Distillation

    Authors' CAST framework redistributes chunk-level teacher advantages across student tokens and anchors unaligned positions, outperforming prior baselines on math, science, and code benchmarks.

    six-benchmark average: 43.87% pass@1 (as reported)

    12 Oct 2026Taobao & Tmall Group of Alibaba, Singapore Management University

14. Research · Evaluation & benchmarks

    Pass@k misses output diversity changes caused by model post-training

    What pass@k Cannot Measure: Evaluating Diversity and Capability Retention after Post-Training

    GRPO and rejection-sampling fine-tuning produced sharply different output distributions despite often similar pass@k scores.

    GSM8K: 0.91 pass@8 (as reported)

    8 Oct 2026Accenture Strategy & Consulting, Vizuara AI Labs

15. Research · Reinforcement learning

    Posterior rewards become indistinguishable for long reasoning traces in LLM training

    Rethinking Probability-Based Reinforcement Learning From Posterior Concentration

    The authors identify a length-dependent failure mode they call Posterior Concentration Phenomenon, and propose a framework to mitigate it.

    MATH-500: 82.9% Avg@2 (as reported)

    7 Oct 2026National University of Singapore, The Hong Kong University of Science and Technology, University of Science and Technology of China

16. Research · Reinforcement learning

    Preserving multiple reasoning strategies prevents late-stage collapse in RLVR

    All Work And No Play Makes Jack a Dull Boy: Understanding and Preventing Catastrophic Strategy Collapse in RLVR

    The authors connect abrupt performance loss to strategy concentration, derive an early-warning metric and test a strategy-balancing training method.

    MATH-500: 86.8% Mean@4 Accuracy (as reported)

    7 Oct 2026Peking University

17. Research · Agents

    More Harness Programs Do Not Necessarily Produce Useful Specialization

    More Programs or More Rolls? Separating Coverage from Specialization in LLM Harnesses

    Controlled comparisons show that apparent gains from diverse LLM harnesses can be matched by repeated runs of identical code.

    MATH-500: 0.01 mean residual correlation (as reported)

    30 Sept 2026The Chinese University of Hong Kong, New Laboratory of Pattern Recognition (NLPR), State Key Laboratory of Multimodal Artificial Intelligence Systems (MAIS)Code