method · first seen 8 Oct 2026
stochastic rounding
A quantization technique that rounds values to adjacent levels with probabilities proportional to distance.
Mentions per week
PapersCodeDiscussion
Rank 124 on 5 Nov 2026 · 7 mentions in 28 days · 7 independent groups · papers · first seen 4 weeks ago · acceleration +1.25
Earliest sources
- Rounding in Preconditioner Space: Redesigning 4-bit AdamW Optimizer-State QuantizationPapers · 8 Oct 2026 · Hanyang Li, Shao Tang, Daniel Thomas Braithwaite, Gregory Dexter, Leonardo Neves…
- What Matters for NVFP4 Training? A Comparative Study of Low-Precision Pretraining RecipesPapers · 17 Oct 2026 · Anjulie Agrusa, Andrei Panferov, Elizabeth Wei, Keith Wyss, Paul Gibbons, Erik S…
- Training an MoE LLM in MXFP4: How We Closed the Gap to BF16Papers · 23 Oct 2026 · Han Wang, Zhitao Wang, Natalia Frumkin, Xuanwu Yin, Alireza Khodamoradi, Dong Li…
- Error Diffusion Replaces Latent Weights: Training Binary-Weight Language Models on a Few Bits of Dithered StatePapers · 25 Oct 2026 · Pierre-Olivier Bonin
- Efficient Quantization Training with Stochastic Rounding via Structured Subspace RandomnessPapers · 27 Oct 2026 · Zeyu Wang, Yiheng Duan, Tao Chen, Donghyun Koo, Sihan Liu, Wenbo Zhang, Chuan He…
- bf16 Rounding Silently Discards Most RL Policy Updates: A Hidden Training-Inference Mismatch and an Unbiased One-Line FixPapers · 29 Oct 2026 · Andrew Gao, William Gao
Latest sources
- Closing the BF16-FP16 Convergence Speed Gap in RL Post-Training with Stochastic RoundingPapers · 10 Nov 2026 · Mahdi Nazemi, Oleg Rybakov, Tijmen Blankevoort, Asit Mishra, Carlo del Mundo
- Lost in Rounding: bf16 Recurrent State Makes Hybrid Reasoning Models Think LongerPapers · 9 Nov 2026 · Yuchuan Weng, Xin Chen, Chi Liu
- Sensitivity-Gated Accumulator Caching for On-Device Learning of Low-Bit Hand NetworksPapers · 7 Nov 2026 · Soroor Ghandali, Kanika Mamgain, Manjot Kaur, Akanksha Shukla, Shashira Guntuka
- Quantizing Recurrent Language Models Is Easier Than You ThinkPapers · 1 Nov 2026 · Philipp Nazari, Neehal Tumma, Daniela Rus, T. Konstantin Rusch
- Where, When, and How Many Bits? Scaling Laws for SGD with Stochastic RoundingPapers · 29 Oct 2026 · Zhichao Wang, Michael W. Mahoney
- bf16 Rounding Silently Discards Most RL Policy Updates: A Hidden Training-Inference Mismatch and an Unbiased One-Line FixPapers · 29 Oct 2026 · Andrew Gao, William Gao
- Efficient Quantization Training with Stochastic Rounding via Structured Subspace RandomnessPapers · 27 Oct 2026 · Zeyu Wang, Yiheng Duan, Tao Chen, Donghyun Koo, Sihan Liu, Wenbo Zhang, Chuan He…
- Error Diffusion Replaces Latent Weights: Training Binary-Weight Language Models on a Few Bits of Dithered StatePapers · 25 Oct 2026 · Pierre-Olivier Bonin
- Training an MoE LLM in MXFP4: How We Closed the Gap to BF16Papers · 23 Oct 2026 · Han Wang, Zhitao Wang, Natalia Frumkin, Xuanwu Yin, Alireza Khodamoradi, Dong Li…
- What Matters for NVFP4 Training? A Comparative Study of Low-Precision Pretraining RecipesPapers · 17 Oct 2026 · Anjulie Agrusa, Andrei Panferov, Elizabeth Wei, Keith Wyss, Paul Gibbons, Erik S…
- Rounding in Preconditioner Space: Redesigning 4-bit AdamW Optimizer-State QuantizationPapers · 8 Oct 2026 · Hanyang Li, Shao Tang, Daniel Thomas Braithwaite, Gregory Dexter, Leonardo Neves…