Pith. sign in

REVIEW 9 cited by

Recipes for Pre-training LLMs with MXFP8

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2506.08027 v2 pith:I5E5PQ66 submitted 2025-05-30 cs.LG cs.AIcs.DC

Recipes for Pre-training LLMs with MXFP8

classification cs.LG cs.AIcs.DC
keywords parameterstensorschoicespre-trainingrepresenttechniquethoseaccuracy
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Using fewer bits to represent model parameters and related tensors during pre-training has become a required technique for improving GPU efficiency without sacrificing accuracy. Microscaling (MX) formats introduced in NVIDIA Blackwell generation of GPUs represent a major advancement of this technique, making it practical to combine narrow floating-point data types with finer granularity per-block scaling factors. In turn, this enables both quantization of more tensors than previous approaches and more efficient execution of operations on those tensors. Effective use of MX-formats requires careful choices of various parameters. In this paper we review these choices and show how MXFP8-E4M3 datatype and a specific number conversion algorithm result in training sessions that match those carried out in BF16. We present results using models with up to 8B parameters, trained on high-quality datasets of up to 15T tokens.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 9 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Four Over Six: More Accurate NVFP4 Quantization with Adaptive Block Scaling

    cs.CL 2025-12 conditional novelty 7.0

    Four Over Six adaptively scales blocks in NVFP4 quantization to smaller FP4 values, making representable value distributions more uniform and reducing quantization error especially for near-maximal values.

  2. Heterogeneity-Aware Microscaling for Efficient Low-Bit LLM Inference

    cs.AR 2026-08 conditional novelty 6.0

    AdaMX puts per-block format and enhancement metadata into spare MX exponent bits, recovering 83% of MXFP4's commonsense accuracy loss at the same 4-bit storage budget.

  3. Stable FP4 Training via Transposition-Invariant Block Quantization

    cs.LG 2026-07 conditional novelty 6.0

    Transposition-invariant 2D block FP4 quantization plus truncation-free scaling and stochastic rounding enables stable end-to-end FP4 LLM training within ~1% of BF16.

  4. SOAP, Muon, and Beyond: Pushing LLM Pretraining Scales

    cs.LG 2026-07 conditional novelty 6.0

    SOAP and Muon, stabilized by per-step QR eigenbasis updates and KL-Shampoo covariance accumulation, beat AdamW on large-batch LLM pretraining up to 100M-token batches.

  5. StoSignSGD: Unbiased Structural Stochasticity Fixes SignSGD for Training Large Language Models

    cs.LG 2026-04 unverdicted novelty 6.0

    StoSignSGD resolves SignSGD divergence on non-smooth objectives via structural stochasticity, matching optimal convex rates and improving non-convex bounds while delivering 1.44-2.14x speedups in FP8 LLM pretraining.

  6. ACRL: Adaptive Control of Training-Inference Discrepancy for Stable Reinforcement Learning

    cs.LG 2026-07 conditional novelty 5.0

    ACRL stabilizes FP8-quantized LLM reinforcement learning by adaptively scaling token-level importance ratios with the measured training-inference discrepancy.

  7. Stochastic Rounding Increases Small Singular Values

    math.NA 2026-05 unverdicted novelty 5.0

    Stochastic rounding lifts clusters of small singular values even in constant aspect ratio matrices, extending its role as a spectral regularizer.

  8. OSP-Next: Efficient High-Quality Video Generation with Sparse Sequence Parallelism, HiF8 Quantization, and Reinforcement Learning

    cs.CV 2026-05 unverdicted novelty 4.0

    OSP-Next reports 83.73% VBench score and up to 2.27x speedup via hybrid sparse attention, SSP parallelism, HiF8 quantization, and Mix-GRPO on diffusion transformers.

  9. A Comprehensive FP8 Training Recipe for Reasoning-Enhanced Language Models

    cs.CL 2025-09 conditional novelty 4.0

    A hybrid-granularity FP8 training recipe achieves near-lossless accuracy versus BF16 on reasoning benchmarks with up to 22% faster training and 14% lower memory.