HBQ: Hierarchical Scaling Block Quantization with Hardware‑Efficiency‑Aware Design for Accurate LLM Inference proposes a new block‑quantization scheme that uses large blocks and low‑overhead significand scaling to balance hardware efficiency and accuracy. The authors demonstrate that larger blocks improve efficiency by amortizing dequantization and accumulation costs, while their SIG scaling compensates for the resulting accuracy loss. Experiments on a 28 nm ASIC accelerator show that HBQ achieves up to 4.6× higher area/energy efficiency than state‑of‑the‑art weight‑only quantization, with 1.5–3.0× speedup and 1.6–3.3× system energy reduction over existing BQ methods.
By Chun-Ting Chen, Dongmin Han, Hangyeol Mun, Jake Hyun, Arnab Raha, Amit Agarwal, Mark Anders, Mohamed Abdelfattah, Jae-sun Seo
The paper introduces MiX, a micro‑inverted‑scaling format that replaces shared exponents with shared mantissas to avoid microscaling collapse in low‑bit vision‑language models. An adaptive dual‑format inference framework (MiX‑MX) maps this format to a custom accelerator, replacing multipliers with shifters. Experiments show 4.5‑bit MiX matches or outperforms NVFP4 accuracy while improving area efficiency by 25 % and delivering 2.3–4.5× speedup with 1.4–2.9× energy savings over the Focus accelerator.
By Yuan Liao, Jae-sun Seo
The paper introduces a new 4‑bit floating‑point (FP4) pretraining approach that pairs E2M1 payloads with unsigned E5M3 block scales, enabling periodic tensor scaling and selective stochastic rounding while eliminating the randomized Hadamard transform. Using this method, the authors pretrained a Nemotron‑H 8B model on nearly 190 billion tokens, achieving lower training and validation losses compared to NVIDIA’s Transformer Engine. The approach also improves inference performance and demonstrates a 21.2 % increase in token throughput when certain optimizations are removed.
By Robert Hu, Carlo Luschi, Paul Balanca
MiX: Micro-Inverted-Scaling for End-to-End Low-Bit Vision-Language Model Acceleration proposes a new quantization format that inverts the traditional microscaling approach by assigning private exponents to each element and a shared mantissa. The adaptive dual-format MiX-MX inference framework maps this format to a custom accelerator, replacing multipliers with shifters. Evaluations show that 4.5-bit MiX matches or surpasses NVFP4 accuracy on multimodal benchmarks while improving area efficiency by 25% and delivering 2.3–4.5× speedup with 1.4–2.9× energy reduction compared to the Focus accelerator.
The paper introduces H-Scale, a lightweight post-processing technique for refining per-group scaling factors in NVFP4 quantized large language models. By using a diagonal second-order proxy from calibration activations, H-Scale selects hardware-valid scales that directly target layer output perturbation rather than just weight reconstruction error. Experiments on mainstream LLMs show that H-Scale improves NVFP4 baselines and brings several variants closer to BF16 performance without adding inference overhead.
By Hao Yu, Zheng Li, Dayiheng Liu, Jianwei Zhang
arXiv:2609.36654v1 Announce Type: new
Abstract: Large language models make weight storage and memory traffic major inference costs, motivating low-precision formats that represent each weight with on...
By Ruiyi Ding, Jie Li, Kang He, Ziyan Liu, Chengru Song, Yuedong Xu, Yuan Cheng
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.
arXiv:2606. 07618v1 Announce Type: cross Abstract: NVFP4 is a recently introduced hardware-supported FP4 format that improves the fidelity of 4-bit quantization through fine-grained block scales.
By Li Lin, Xiaojun Wan
The paper presents a 4‑bit quantization recipe, Minima: NVFP4 W4A4, that fully quantizes all linear layers—including the Gated DeltaNet (GDN) recurrent blocks—of the 27‑billion‑parameter Qwen3.8 LLM. Across a suite of benchmarks (perplexity, MMLU‑Pro, GSM8K, AIME'25, GPQA‑Diamond, LiveCodeBench, and RULER retrieval), the quantized model matches BF16 performance within seed noise while being 17.5 GiB in size and 14–19 % faster at prefill. The authors attribute this success to four mechanisms: block‑scaling of residuals, robust gate projections, the delta‑rule recurrence’s noise‑plateau behavior, and the per‑token quantization cost’s dilution over long contexts.
By Sergii Kozyrev, Davyd Maiboroda
arXiv:2606. 06521v1 Announce Type: cross Abstract: FP8 (E4M3) acceleration for attention computation offers significant throughput gains, but the 3-bit mantissa introduces precision challenges when the softmax probability matrix P is cast to FP8 before the P*V matrix multiplication.
By Reed Lau
arXiv:2607. 02893v1 Announce Type: new Abstract: Low-bit quantization shrinks language models but treats precision as a single global hyper-parameter: every weight uses the same bit-width.
By Hamish Ogilvy
arXiv:2609.37887v1 Announce Type: new
Abstract: Activation and key-value cache precision change what a quantized language model computes without altering its stored weights. Direct weight-code bounds...
By Arian Eamaz, Mojtaba Soltanalian