method · first seen 5 Oct 2026
NVFP4 quantization
Also: MX, NVFP4, NVIDIA FP4, 4-bit floating point, microscaling formats, FP4 microscaling formats, NVIDIA 4-bit floating point
A 4-bit floating-point number format with block scaling for quantizing neural network weights and activations on NVIDIA hardware.
NVFP4 is NVIDIA's 4-bit floating-point format (E2M1 values with a shared scale per 16-element block) that runs natively on Blackwell tensor cores. The sources show work moving past inference-only quantization toward training with it. Several groups apply it to RL post-training and report learner-sampler mismatch and the fixes they built for it. One comparative study, co-authored by Dan Alistarh, evaluates NVFP4 pretraining recipes at 8B and 30B MoE scale. Other papers cover smoothing and scale optimization, FP4 KV caches, LoRA fine-tuning on consumer GPUs, and per-tile format selection. Hobbyist repos ship vLLM and SGLang serving recipes for DGX Spark and RTX cards.
If NVFP4 training and RL prove stable at scale, 4-bit precision could become the default for both serving and training on current NVIDIA hardware, roughly doubling throughput over FP8. Caveats remain, such as slower acquisition of rare facts during FP4 pretraining.
Analyst note (emerging), 17 Nov 2026.
Mentions per week
PapersCodeDiscussion
Rank 4 on 16 Nov 2026 · 8 mentions in 28 days · 8 independent groups · papers → code → discussion · first seen 6 weeks ago · acceleration +1.83
Earliest sources
- TRIAGE: Direction-Aware Mismatch Stabilization of Native NVFP4 Reinforcement LearningPapers · 5 Oct 2026 · Zhen Li, Shuai Zhang, Yanggan Gu, Yiming Zhang, Yang Yu, Mingfa Feng, Congkai Xi…
- OSFP4: Joint Optimization of Diagonal Smoothing and Block Scales for NVFP4 QuantizationPapers · 6 Oct 2026 · Neriah Ben David, Ori Meir, Or Ordentlich
- miniTE: Diagnosing and Bridging the Execution Gap in Native NVFP4 Reinforcement LearningPapers · 7 Oct 2026 · Zhen Li, Mingfa Feng, Yiming Zhang, Yang Yu, Changhai Zhou, Pengxiang Li, Dongku…
- meshapplied/goetheCode · 8 Oct 2026 · meshapplied · 69 stars
- salient-data/qwen38-next-dgx-recipeCode · 8 Oct 2026 · salient-data · 35 stars
- DualQuant: Dual-scale Calibration-free Weight-and-Activation Quantization for LLMsPapers · 9 Oct 2026 · Mahsa Salmani, Sayeh Sharify, Hesham Mostafa
Latest sources
- MiSTiQ: MicroScaling Trellis Quantization for Extreme KV Cache CompressionPapers · 16 Nov 2026 · Ofir Gordon, Lior Dikstein, Paul Balanca, Idan Achituve, Douglas Orr, Hai Victor…
- Gradient-Guided Hardware-Native MIXFP4Papers · 12 Nov 2026 · Jian-Jia Chen, Brian J. Chan, Wen-Hung Lee, Yi-Ting Li, Po-Shen Chen, Kuan-Te Ku…
- What the Loss Curve Misses: FP4 Pretraining Slows the Acquisition of Rare FactsPapers · 9 Nov 2026 · Chang Lu, Mengjiao Zhang
- AutoEMatic: Runtime-Adaptive Mixed-Precision LLM Inference via Automatic Exponent-Mantissa RefinementPapers · 6 Nov 2026 · Akshat Ramachandran, Reena Elangovan, Rangharajan Venkatesan, Charbel Sakr, Tijm…
- BitDecoding-FP4: Native FP4 KV Cache Decoding via Algorithm-System Co-DesignPapers · 6 Nov 2026 · Dayou Du, Woon Yee Ng, Guangming Sheng, Nick Lau, Jianyi Cheng, Luo Mai
- FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel CompensationPapers · 3 Nov 2026 · Yining Wei, Tianyuan Wu, Gu Gong, Yuechen Tao, Yinghao Yu, Xiaowen Chu, Wei Wang
- xl03-bit/sm120-nvfp4-trainingCode · 2 Nov 2026 · xl03-bit · 103 stars
- Running Qwen3.8-Flash-Next NVFP4 at 184 tok/s on a single RTX 6000 ProDiscussion · 26 Oct 2026 · matteohorvath · 29 points
- PACT: NVFP4 Data-Flow LoRA Fine-Tuning on Consumer GPUsPapers · 18 Oct 2026 · Lei Gao, Hossein Entezari Zarch, Chaoyi Jiang, Murali Annavaram
- 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…
- DualQuant: Dual-scale Calibration-free Weight-and-Activation Quantization for LLMsPapers · 9 Oct 2026 · Mahsa Salmani, Sayeh Sharify, Hesham Mostafa
- salient-data/qwen38-next-dgx-recipeCode · 8 Oct 2026 · salient-data · 35 stars
Mentioned together with
In the ledger
- 2026-W47 · call #3 · open