FoldAttention: Declared-Reference Softmax for Fast Decode and Deterministic Backward

arXiv:2609.33410 · cs.LG, cs.DC · Submitted 2026-09-27 · Read on arXiv

cs.LG, cs.DC

Submitted: 2026-09-27

Updated: 2026-09-27

Code: https://github.com/srimanachanta/fold-attention

Terminology

• Decode — 61× in this paper · explained in 2 episodes
• Accuracy — 11× in this paper · explained in 1 episode
• Core — 8× in this paper · explained in 2 episodes
• KV cache — 5× in this paper · explained in 3 episodes
• Verification — 5× in this paper · explained in 2 episodes
• Transformer — 4× in this paper · explained in 2 episodes
• Benchmark — 3× in this paper · explained in 2 episodes
• Sparsity — 3× in this paper · explained in 4 episodes
• Ablation — 2× in this paper · explained in 1 episode
• Adam — 2× in this paper · explained in 2 episodes
• Quantization — 2× in this paper · explained in 12 episodes
• Attention Mechanism — 1× in this paper · explained in 3 episodes
• Calibration — 1× in this paper · explained in 7 episodes
• Reinforcement Learning — 1× in this paper · explained in 4 episodes
• Sparse Attention — 1× in this paper · explained in 5 episodes
• Speculative Decoding — 1× in this paper · explained in 2 episodes
• Transformer Models — 1× in this paper · explained in 2 episodes

Sources

• FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning
• The Llama 3 Herd of Models
• Hydragen: High-Throughput LLM Inference with Shared Prefixes
• Faster Than Flash: Exploiting Attention Sparsity for Efficient Long-Context Decoding
• FP8 Formats for Deep Learning
• Online normalizer calculation for softmax
• gpt-oss-120b & gpt-oss-20b Model Card
• DASH: Deterministic Attention Scheduling for High-throughput Reproducible LLM Training
• SparQ Attention: Bandwidth-Efficient LLM Inference
• Lean Attention: Hardware-Aware Scalable Attention Mechanism for the Decode-Phase of Transformers
• FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision
• VFA: Relieving Vector Operations in Flash Attention with Global Maximum Pre-computation
• ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools
• CoSA: Accelerating Long-Context Inference via Proxy-Kernel Co-Designed Sparse Attention
• Qwen3 Technical Report
• FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving
• BLASST: Dynamic BLocked Attention Sparsity via Softmax Thresholding
• FlashAttention-4: Algorithm and Kernel Pipelining Co-Design for Asymmetric Hardware Scaling
• Deterministic Inference across Tensor Parallel Sizes That Eliminates Training-Inference Mismatch
• Diagnosing Training Inference Mismatch in LLM Reinforcement Learning via a Zero-Mismatch Reference

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks