Pith. sign in

REVIEW 26 cited by

Microscaling Data Formats for Deep Learning

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 2310.10537 v3 pith:S573HYGJ submitted 2023-10-16 cs.LG cs.AI

classification cs.LGcs.AI
keywords formatsdatatrainingaccuracydeepfrictionlearningmicroscaling
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Narrow bit-width data formats are key to reducing the computational and storage costs of modern deep learning applications. This paper evaluates Microscaling (MX) data formats that combine a per-block scaling factor with narrow floating-point and integer types for individual elements. MX formats balance the competing needs of hardware efficiency, model accuracy, and user friction. Empirical results on over two dozen benchmarks demonstrate practicality of MX data formats as a drop-in replacement for baseline FP32 for AI inference and training with low user friction. We also show the first instance of training generative language models at sub-8-bit weights, activations, and gradients with minimal accuracy loss and no modifications to the training recipe.

Discussion (0). Sign in to comment.

Forward citations

Cited by 26 Pith papers

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

  1. When Does Disaggregation Pay? Simulating Prefill--Decode--Attention--FFN Specialization for Agentic LLM Inference

    cs.DC 2026-08 conditional novelty 7.0 of 10

    A simulator predicts that four-way prefill/decode/attention/FFN disaggregation beats unified serving on prefill-heavy agentic workloads by up to 2.06x, but only with stage-specialized custom NPUs and rich enough hardw...

  2. One QK Channel, Many Sources: Guarding Low-Precision Attention Collapse

    cs.LG 2026-08 conditional novelty 7.0 of 10

    Different low-precision errors converge on the same query-key spectral runaway, entry is gated by temporal sign-coherence, and a dormant query-key normalization guard contains it.

  3. Jack of All Scales: A Versatile FPGA Tensor Block for MXFP Precisions

    cs.AR 2026-07 conditional novelty 7.0 of 10

    Targeted DSP tensor-mode changes enable native MXFP4/MXFP6/E4M3 support on Agilex-5-like FPGAs, with a 36% block-area cost and 4.2x average systolic-array throughput gain over baseline mapping strategies.

  4. Full-Stack FP4: Stable LLM Pretraining with Quantized Projections, Optimizers, and Attention

    cs.LG 2026-07 conditional novelty 7.0 of 10

    Full-stack NVFP4 LLM pretraining is stable with a 1.47% loss gap to BF16 via LoRA-SVD projections, transformed AdamW/Root optimizers, and mixed-precision attention.

  5. Characterization and Mitigation of Training Instabilities in Microscaling Formats

    cs.LG 2025-06 conditional novelty 7.0 of 10

    Block-scaled MX low-precision training is unstable because quantization of tightly clustered layer-norm weights and some activations injects multiplicative gradient bias, and this can be fixed by keeping activations i...

  6. FP4 All the Way: Fully Quantized Training of LLMs

    cs.LG 2025-05 conditional novelty 7.0 of 10

    A 7-billion parameter LLM trained almost entirely in 4-bit float precision matches a standard 16-bit baseline after a brief quantization-aware fine-tuning phase.

  7. MXAttention: Data-Free Optimal Scaling and Pre-Normalization Quantization for MXFP4 Attention

    cs.LG 2026-07 conditional novelty 6.5 of 10

    A closed-form MXFP4 scale boundary Qmax=7.25 plus pre-normalization softmax quantization recovers nearly all FP16 video quality lost by naive MXFP4 attention.

  8. dMX: Differentiable Mixed-Precision Assignment for Low-Precision Floating-Point Formats

    cs.LG 2026-06 unverdicted novelty 6.5 of 10

    dMX learns continuous per-layer MXFP bit-width offsets with temperature annealing and target-aware regularization, yielding Pareto-better mixed-precision LLMs than uniform or KL-heuristic baselines.

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

    cs.AR 2026-08 conditional novelty 6.0 of 10

    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.

  10. Studying quantization trade-offs for efficient inference deployment in machine translation

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Quantized Hy-MT2 models stay accurate at long context, but quantized EuroLLM 9B/22B models collapse (up to ~60% chrF++ drop) while W4A8/W8A8 plus 200–400-token chunking improves serving throughput.

  11. LightRot: A Light-Weighted Rotation Scheme and Architecture for Accurate Low-Bit Large Language Model Inference

    cs.AR 2026-07 conditional novelty 6.0 of 10

    LightRot uses grouped local rotation plus outlier alignment to make 4-bit LLaMA inference accurate and cheap, claiming 27.4 TOPS/W on a 28nm accelerator.

  12. GyRot: Leveraging Hidden Synergy between Rotation and Fine-grained Group Quantization for Low-bit LLM Inference

    cs.AR 2026-07 conditional novelty 6.0 of 10

    Rotation and fine-grained group quantization can work together if rotation spans several quantization groups and outlier channels are permuted onto harmonic Hadamard rows, enabling 4-bit LLM inference with integer-onl...

  13. Stable FP4 Training via Transposition-Invariant Block Quantization

    cs.LG 2026-07 conditional novelty 6.0 of 10

    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.

  14. MXSens: Sensitivity-Aware Mixed-Precision Quantization for Efficient LLM Inference

    cs.LG 2026-07 conditional novelty 6.0 of 10

    MXSens allocates 8-bit precision to the 32 most sensitive columns per layer, 6-bit to moderately sensitive columns, and 4-bit elsewhere in MXINT, improving WikiText-2 perplexity over prior 4-bit LLM quantization methods.

  15. CANN Bench: Benchmarking Agent Generated Kernels against Real NPU and Algorithmic Limits

    cs.AI 2026-07 conditional novelty 6.0 of 10

    CANN Bench is an Ascend-NPU benchmark with 53 operators and 1,060 released test cases that scores agent-generated kernels on compilation, correctness, and hardware-anchored performance.

  16. DynamiQ: Accelerating Gradient Synchronization using Compressed Multi-hop All-reduce

    cs.LG 2026-02 conditional novelty 6.0 of 10

    DynamiQ compresses gradients for multi-hop all-reduce with per-block bit-width allocation, block reordering, and fused recompression, matching BF16 accuracy at ~5 bits per coordinate and improving time-to-accuracy by ...

  17. ARCQuant: Boosting NVFP4 Quantization with Augmented Residual Channels for LLMs

    cs.LG 2026-01 conditional novelty 6.0 of 10

    By appending quantized residual channels to NVFP4 activations and duplicating the matching weights, ARCQuant reaches W4A8-level accuracy while keeping a single unified 4-bit GEMM.

  18. SeVeDo: A Heterogeneous Transformer Accelerator for Low-Bit Inference via Hierarchical Group Quantization and SVD-Guided Mixed Precision

    cs.LG 2025-12 conditional novelty 6.0 of 10

    SeVeDo achieves 12.7–13.8 TOPS/W in 28nm for ViT-Base and Llama2-7B by combining SVD-based low-rank decomposition, hierarchical group quantization, and SVD-guided mixed precision.

  19. Hybrid Systolic Array Accelerator with Optimized Dataflow for Edge Large Language Model Inference

    cs.AR 2025-07 conditional novelty 6.0 of 10

    A 28nm hybrid systolic array accelerator with MXINT4 quantization and fused RMSNorm/RoPE units reports 247/117 token/s/mm2 running RetNet 1.3B, claiming over 2.45x/13.5x area-efficiency gains over prior edge LLM accelerators.

  20. Recipes for Pre-training LLMs with MXFP8

    cs.LG 2025-05 conditional novelty 6.0 of 10

    MXFP8-E4M3 with round-up scale computation matches BF16 pre-training accuracy for LLMs up to 8B parameters, enabling 2x faster training on Blackwell.

  21. FPQVAR: Floating Point Quantization for Visual Autoregressive Model with FPGA Hardware Co-design

    cs.CV 2025-05 conditional novelty 6.0 of 10

    A post-training floating-point quantization framework with grouped Hadamard rotation and learned smoothing brings 4-bit visual autoregressive image generation to near-FP16 quality, plus a matching FPGA accelerator.

  22. WINT: A Novel Weighted Integer Representation with Improved Error Characteristics

    cs.AR 2026-07 accept novelty 5.5 of 10

    WINT (M imes2^E) with 2–3 exponent bits gives both larger range and lower mean relative error than same-width integers for bit widths of 12 and above.

  23. FOCUS: FP4 Optimization via Coupled-Relaxation and Dual-Granularity Scaling

    cs.AI 2026-08 conditional novelty 5.0 of 10

    FOCUS relaxes the tie between quantization and dequantization scales in FP4 formats, learning a full-precision coefficient and sub-block scales to find better FP4 weight assignments.

  24. TorchAO: PyTorch-Native Training-to-Serving Model Optimization

    cs.LG 2025-07 conditional novelty 5.0 of 10

    TorchAO delivers a PyTorch-native, end-to-end workflow for FP8 training, QAT, PTQ, and sparsity, with benchmarked speedups and production use in quantized Llama releases.

  25. Beyond the Buzz: A Pragmatic Take on Inference Disaggregation

    cs.DC 2025-06 conditional novelty 5.0 of 10

    Disaggregated serving (separate prefill and decode GPU pools) expands the throughput-interactivity Pareto frontier mainly for prefill-heavy workloads and models larger than about 10B parameters, provided the prefill-t...

  26. How to keep pushing ML accelerator performance? Know your rooflines!

    cs.AR 2025-05 conditional novelty 4.0 of 10

    A JSSC survey organizing ML accelerator techniques around two roofline curves, one for throughput and one for energy efficiency, with the insight that their bottleneck knees often sit at different arithmetic intensities.

Pith tools