UE5M3 FP4 Block Scaling for Stable Language Model Pretraining

• URL: http://arxiv.org/abs/2609.02846v1
• Date: Wed, 02 Sep 2026 17:32:07 GMT
• Title: UE5M3 FP4 Block Scaling for Stable Language Model Pretraining
• Abstract summary: We pretrain a model for nearly 190 billion tokens using NVIDIA's Transformer Engine nv recipe.<n>Compared with Transformer Engine nv, the proposed block-16 recipe finishes with lower final-window training loss and, under their respective quantized-inference policies, lower validation loss measured as held-out negative log-likelihood.<n>A native nv execution ablation that jointly removes RHT and the BF16 final-block exemption increases measured model-body token throughput by 21.2%.
• Score: 6.963061311516306
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Stable 4-bit floating-point (FP4) pretraining is difficult because the E2M1 payload represents only a narrow range of magnitudes. NVIDIA's Transformer Engine \nv{} recipe addresses this with current-tensor scaling, a randomized Hadamard transform (RHT), and bfloat16 (BF16) final layers, adding work outside the FP4 matrix multiplications. We instead pair E2M1 payloads with unsigned E5M3 (\ue{}) block scales. Their wider range permits periodic tensor scaling, while our recipe applies selective stochastic rounding to backward gradients, omits RHT, and uses FP4 in all eligible internal linears. We pretrain a Nemotron-H 8B model for nearly 190 billion tokens. Compared with Transformer Engine \nv{}, the proposed block-16 recipe finishes with lower final-window training loss and, under their respective quantized-inference policies, lower validation loss measured as held-out negative log-likelihood. Its quantized-inference downstream point estimates are also higher on all three reported aggregates. A native \nv{} execution ablation that jointly removes RHT and the BF16 final-block exemption increases measured model-body token throughput by 21.2\%. These results demonstrate end-to-end software-emulated \uefp{} pretraining with a simpler recipe and motivate native support for \ue{} block scaling.

Related papers

• Stable FP4 Training via Transposition-Invariant Block Quantization [14.373963304887754]
  In conventional 1D block quantization, forward and backward passes assign di erent scaling factors to the same values after transposition.<n>We propose a low-precision training framework based on 2D block FP4 quantization, which enforces transposition-invariant scaling and preserves consistency between forward and backward computations.<n>Our approach achieves stable end-to-end FP4 training and closely matches BF16 performance, with less than 1.3% degradation in perplexity and downstream accuracy.
  arXiv  Detail & Related papers  (2026-07-27T18:03:34Z)
• Hidden Decoding at Scale: Latent Computation Scaling for Large Language Models [75.80275843320511]
  We study whether an existing backbone can keep improving by allocating more computation to each token while leaving the Transformer backbone fixed.<n>We propose Hidden Decoding, a sequence-length scaling method applied during continued pretraining.
  arXiv  Detail & Related papers  (2026-07-09T07:37:59Z)
• FourTune: Towards Fully 4-Bit Efficient Post-Training for Diffusion Models [50.66675248468456]
  FourTune is an efficient post-training framework for diffusion models based on an end-to-end W4A4G4 paradigm.<n>FourTune reduces memory overhead by 2.25$times$ and increases end-to-end training throughput by 2.27$times$ compared to BF16 LoRA.
  arXiv  Detail & Related papers  (2026-07-07T00:34:30Z)
• Full-Stack FP4: Stable LLM Pretraining with Quantized Projections, Optimizers, and Attention [34.07319488267708]
  Recent NVFP4 pretraining methods mainly target transformer linear layers, leaving computation states, arithmetic and attention underexplored in 4-bit pipelines.<n>This critical gap blocks stable full-stack 4-bit pretraining, as the three core modules exhibit unique numerical failure patterns.<n>We propose Full-Stack FP4, the first complete NVFP4 pretraining framework resolving all three stability bottlenecks.
  arXiv  Detail & Related papers  (2026-07-05T17:31:36Z)
• SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference [4.219937026460372]
  SharQ bridges activation sparsity and FP4 quantization through an online sparse--dense decomposition.<n>A sparse FP4 GEMM processes the backbone while a dense FP4 GEMM compensates for both mask-induced activation loss and sparse-path quantization error.
  arXiv  Detail & Related papers  (2026-06-25T04:19:04Z)
• Rethinking Shrinkage Bias in LLM FP4 Pretraining: Geometric Origin, Systemic Impact, and UFP4 Recipe [31.895766254664167]
  We propose a uniform 4-bit training recipe that applies RHT to all three training GEMMs while restricting rounding to dY alone.<n>On Dense 1.5B, MoE 7.9B, and MoE 124B long-run pretraining, UFP4 consistently achieves lower BF16-relative loss degradation than strong E2M1-based baselines.<n>Our results suggest that future accelerators should support E1M2/INT4-style uniform 4-bit grids as first-class training primitives alongside E2M1.
  arXiv  Detail & Related papers  (2026-06-18T15:40:51Z)
• N-vium: Mixture-of-Exits Transformer for Accelerated Exact Generation [68.47358899451255]
  N-vium is a mixture-of-exits transformer that partially parallelizes computation across depth on standard hardware.<n>N-vium attaches prediction heads at multiple depths and defines the next-token distribution as a learned mixture over these exits.
  arXiv  Detail & Related papers  (2026-05-13T08:46:17Z)
• Normalized Architectures are Natively 4-Bit [49.13186675123547]
  nGPT, an architecture that constrains weights and hidden representations to the unit hypersphere, is inherently more robust to low-precision arithmetic.<n>We validate this approach on both a 1.2B dense model and hybrid (Mamba-Transformer) MoE models of up to 3B/30B parameters.
  arXiv  Detail & Related papers  (2026-05-07T11:54:07Z)
• Speeding Up MACE: Low-Precision Tricks for Equivarient Force Fields [51.95157731126864]
  Machine-learning force fields can deliver accurate molecular dynamics (MD) at high computational cost.<n>This thesis aims to make MACE cheaper and faster by identifying computational bottlenecks and evaluating low-precision execution policies.
  arXiv  Detail & Related papers  (2025-10-23T14:02:34Z)
• FP4 All the Way: Fully Quantized Training of LLMs [26.195547788434908]
  We demonstrate fully quantized training (FQT) of large language models (LLMs) using predominantly 4-bit floating-point (FP4) precision.<n>We investigate key design choices for FP4, including block sizes, scaling formats, and rounding methods.
  arXiv  Detail & Related papers  (2025-05-25T12:14:25Z)
• SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models [61.474101404805545]
  Diffusion models can generate high-quality images, but as they scale, rising memory demands and higher latency pose deployment challenges.<n>We propose SVDQuant, a new 4-bit quantization paradigm to overcome this limitation.<n>We reduce the memory usage for the 12B FLUX.1 models by 3.5$times$, achieving 3.0$times$ speedup over the 4-bit weight-only quantization (W4A16) baseline.
  arXiv  Detail & Related papers  (2024-11-07T18:59:58Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.