GSM8K

Timeline

1. Oct 19, 2026 · Opinion / analysis · 1 source
   Qwen 3.8 Flash Next optimized for two CMP 170hx GPUs
   I've been working on optimizing Qwen 3.8 Flash Next on my local setup with two CMP 170HX cards and finally got to a place where I feel like the results are worth sharing.
2. Oct 16, 2026 · Research paper · 1 source
   Trajectory-Triggered Sparse Collaborative Decoding
   We study sparse collaborative decoding, where a draft model generates most tokens and a target model is invoked only for short corrective segments.
3. Oct 16, 2026 · Research paper · 1 source
   ReLaDrift: Recurrent Language Generation via Drifting for Test-Time Scaling
   We propose Recurrent Language Generation via Drifting (ReLaDrift), which uses a shared Drifting Model to generate an initial draft from noise and iteratively refine it in a continuous latent space.
4. Oct 16, 2026 · Research paper · 2 sources
   Towards Looped Models Done Right, Part I: Topology, Input Injection, Recurrent-State Design
   Looped language models reuse physical Transformer blocks across logical depth, but established architectures entangle recurrence placement, input injection, and recurrent state organization.
5. Oct 16, 2026 · Research paper · 1 source
   Sep-Latent: Separator Tokens as Natural Compression Anchors for Implicit Reasoning
   In this paper, we propose Sep-Latent, an adaptive implicit reasoning framework that utilizes separator tokens as structural semantic anchors.
6. Oct 16, 2026 · Research paper · 1 source
   Next Token or Vector? Learning How Far to Decode
   We propose CADENCE (Context-Adaptive DEcoding), a generation method that learns both what to draft and how far to draft.
7. Oct 15, 2026 · Research paper · 1 source
   Diverse Reasoning Paths Matter: Symbolic Equivalence Augmentation for Enhancing RL Exploration
   Reinforcement learning with verifiable rewards (RLVR) has emerged as a powerful paradigm for improving the reasoning capabilities of LLMs. However, the role of supervised fine-tuning (SFT) data in enabling effective downstream RL remains underexplored.
8. Oct 15, 2026 · Opinion / analysis · 1 source
   capy-flash: Qwen3.8-Flash-Next (125B MoE) in full NVFP4 on two 16 GB consumer GPUs
   capy-flash is the evolution of my ninfer-tp2 engine.
9. Oct 15, 2026 · Research paper · 1 source
   TIPS: A Tiny Inference-time Policy for Scaling in Diffusion Language Models
   Masked diffusion language models expose an inference-time choice: which positions become visible after each denoising pass.
10. Oct 15, 2026 · Research paper · 1 source
    The Density of Thought: Predicting and Steering Reasoning in Large Language Models
    We investigate whether variation in information density across chain-of-thought (CoT) reasoning steps predicts answer correctness and can serve as a controllable inference-time target.
11. Oct 15, 2026 · Research paper · 1 source
    KBS: A Korean Benchmark Suite for Foundation Model Evaluation
    We present a Korean Benchmark Suite (KBS) comprising Ko-ARC, Ko-GSM8K, Ko-EQ-Bench, Ko-WinoGrande, Ko-LAMBADA, and Ko-IFEval, covering six capabilities across 9,396 items.
12. Oct 14, 2026 · Research paper · 1 source
    INTACT: Inducing Latent Reasoning on a Frozen Backbone
    We use a small external encoder to induce latent reasoning in a frozen backbone model.
13. Oct 14, 2026 · Research paper · 1 source
    Target-Token Asymmetry Can Masquerade As Depth Specialization
    Layer-ablation studies often perturb early or late layers of a language model, compare the damage to prediction on two corpora, and read a larger early-layer effect on one corpus as early-layer specialization for its task.
14. Oct 13, 2026 · Research paper · 1 source
    TRACER: Trace-Regularized Training Against Indirect Prompt Injection
    We propose TRACER (Trace-Regularized Adversarial Channel Exposure with Reward-shaping), a channel-aware adversarial training method that samples presentation wrappers and combines an answer-anchored, evidence-gated reward with ATRC, which aligns attacked reasoning with clean evidence when an injection is invalid.
15. Oct 13, 2026 · Research paper · 1 source
    A Few More Tokens Ahead: Reinforcing Parallelism in Diffusion LLMs
    Parallel generation is widely regarded as an inherent advantage of diffusion large language models.
16. Oct 13, 2026 · Research paper · 1 source
    LoOPSD: On-Policy Self-Distillation with Looped Teachers
    To address this, we introduce LoOPSD, which searches over a model's layer execution order to construct a stronger teacher and distills its advantage into the student.
17. Oct 13, 2026 · Research paper · 1 source
    Dynamic K-Head: Difficulty-Supervised Budget Selection for Continuous Latent Reasoning
    We introduce Dynamic K-Head, a question-conditioned, pre-rollout budget selector.
18. Oct 13, 2026 · Research paper · 1 source
    Towards Green AI: Evaluating Energy Efficiency for Quantized LLM Inference
    We present the first unified evaluation of weight-only quantization strategies for LLM inference across varying input/output lengths and GPUs, highlighting differences in energy efficiency in addition to accuracy degradation and runtime.
19. Oct 12, 2026 · Research paper · 1 source
    PIG: Parallel Inconsistency Guidance for Accelerating Diffusion Language Models
    Diffusion language models accelerate generation by decoding multiple masked positions per forward pass.
20. Oct 12, 2026 · Research paper · 1 source
    RL Post-Training Diffusion Language Models Along Their Own Generation Paths
    Diffusion language models are a promising and more efficient alternative to autoregressive models.
21. Oct 12, 2026 · Research paper · 1 source
    Soft Teacher Forcing
    We introduce Soft Teacher Forcing (Soft-TF), a simple training algorithm that combines teacher-forced embeddings with soft tokens through random convex combinations.
22. Oct 12, 2026 · Research paper · 1 source
    ACED: Reference-Anchored Aggregation of Natural-Language and Executable Evidence for Mathematical Reasoning
    Thus we propose ACED, a reference-anchored framework that treats the self-consistency answer as the default and executable outputs as supporting evidence rather than interchangeable votes.
23. Oct 12, 2026 · Research paper · 1 source
    Intent-Triggered Anchoring for Multi-Token Format Units in Parallel Decoding of Diffusion Language Models
    In parallel decoding, diffusion language models commit several tokens per forward pass, so co-committed tokens cannot see each other, breaking multi-token format units that parsers need.
24. Oct 11, 2026 · Opinion / analysis · 1 source
    I tested different Qwen 3.8 27B quants
    There is a lot of discussion which quant to use.
25. Oct 11, 2026 · Research paper · 1 source
    GuidanceNFT: NFT-Based Reward Fine-Tuning of Guided Diffusion Models
    DiffusionNFT (NFT) fine-tunes diffusion models toward task-specific rewards, but optimizing these rewards does not guarantee fidelity to the full prompt.
26. Oct 11, 2026 · Research paper · 1 source
    Truncation Samplers Are Nested: Comparing Decoding Rules Is a Measurement Problem
    Settling one comparison between two truncation samplers has cost six thousand A100-hours, and nothing says in advance how far apart two samplers can be.
27. Oct 11, 2026 · Research paper · 1 source
    SpecFit: Not All Draft Tokens Cost the Same in Offloaded MoE Inference
    Mixture-of-experts (MoE) models activate only a few experts per token, but when their expert weights are offloaded to host memory, host-to-device (H2D) expert transfers enter the critical path of decoding.
28. Oct 10, 2026 · Research paper · 1 source
    Running Fewer Loops Removes Refusal Behaviour in Recurrent-Depth Language Models
    A recurrent-depth language model reuses one block of weights several times in a row.
29. Oct 10, 2026 · Research paper · 1 source
    To Defer or Not to Defer? Online Learning with LLMs and Congested Experts
    We propose KL-PBD, a learning algorithm that uses Kullback--Leibler lower confidence bounds on the unknown reliability parameters to guide deferral decisions.
30. Oct 9, 2026 · Research paper · 1 source
    RUBATO: Decoupled Clocks Allow Full-Duplex Models to Think as Fast as They Can
    Recent Speech Full-Duplex (FDX) models are capable of engaging in naturalistic conversation by listening and speaking at the same time.

Often appears with