Metraipolis › archive › dc_bc3d404b2ec4
FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel Compensation — arXiv:2611.04887
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.04887
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-04T18:35:18.610Z
- SHA-256 of the archived text
98106dea6e5c68a0830f273ed5a078644e0c8d07bd9979f35645db1bfe1754f7- Stable URL of this page
https://metraipolis.com/documento/dc_bc3d404b2ec4
How to cite
FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel Compensation — arXiv:2611.04887 — arXiv (en), https://arxiv.org/abs/2611.04887 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bc3d404b2ec4 (Metraipolis archive dc_bc3d404b2ec4, sha256 98106dea6e5c68a0830f273ed5a078644e0c8d07bd9979f35645db1bfe1754f7)
BibTeX
@misc{dcbc3d404b2ec4,
title = {FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel Compensation — arXiv:2611.04887},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.04887},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bc3d404b2ec4; sha256 98106dea6e5c68a0830f273ed5a078644e0c8d07bd9979f35645db1bfe1754f7},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04887. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Reinforcement learning (RL) post-training increasingly targets low-bit deployment, with NVFP4 weight-and activation quantization (W4A4) becoming the state-of-the-practice format. However, integrating NVFP4 into RL degrades rollout quality and amplifies cross-engine training-inference mismatch (TIM), destabilizing training. We show that these failures are concentrated rather than uniform - severely mismatched tail tokens concentrate in a few outlier sequences, while policy weights remain largely intact and quantization error concentrates in a few high-variance activation channels. Guided by this, we present FourSight, an algorithm-system co-design with two targeted solutions: TIM-aware sequence filtering that excludes drifted trajectories from policy updates, and sparse precision compensation that keeps only top-variance channels in BF16 while retaining native NVFP4 elsewhere. We further design a fused SpliceGEMM kernel with specialized prefill and decode datapaths to minimize overhead. FourSight sustains near-BF16 accuracy and strong FP4 deployment quality across models and engines, while improving step-to-accuracy by 1.76-2.60x and staying within 2.3% of FP4 step time up to 120B-scale models. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2611.04887 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-11-04T18:32:37.348Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).