Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM

• URL: http://arxiv.org/abs/2609.04098v1
• Date: Thu, 03 Sep 2026 17:04:26 GMT
• Title: Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM
• Abstract summary: Minima: NVFP4 W4A4 on all 496 linear layers, GDN included.<n>A four-part mechanism study explains why NVFP4's 16-element block scaling localizes the residual stream's extreme outliers.<n>A mechanistic account of why the recurrent half of a hybrid LLM is the easy half to quantize.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Hybrid LLMs pair softmax attention with linear-attention layers such as Gated DeltaNet (GDN), whose recurrent state summarizes the context in fixed size. Early community 4-bit quantizations of Qwen3.8-27B (48 GDN layers, 16 attention layers) left the GDN block in 8- or 16-bit precision -- especially its decay and write-strength gates -- on the intuition that errors in a recurrence accumulate over long contexts. We test that intuition by building Minima: NVFP4 W4A4 on all 496 linear layers, GDN included. Across perplexity at 4K/32K, MMLU-Pro, GSM8K, AIME'25, GPQA-Diamond, LiveCodeBench, and RULER retrieval to 64K, Minima matches BF16 within seed noise (5-task average -0.52) while being the smallest (17.5 GiB) and fastest-prefill (+14-19%) recipe we compare, and its 32K perplexity gap shrinks with position. A four-part mechanism study explains why: (i) NVFP4's 16-element block scaling localizes the residual stream's extreme outliers, equalizing activation error across layer roles; (ii) the supposedly fragile gate projections are the least sensitive -- softplus/exponential and sigmoid parameterizations compress ~11% GEMM error to ~2% output error; (iii) the delta-rule recurrence holds injected noise at a flat plateau over 32K tokens and forgets a state impulse within hundreds of steps, because each write overwrites the state along the current key direction; (iv) the per-token quantization cost washes out with context instead of compounding. We also repair a global-scale mismatch that arises when per-module-calibrated NVFP4 checkpoints are served by kernels that fuse those modules into one GEMM, and show calibrated FP8 KV-cache scales are performance-free. The result: a practical recipe -- quantize everything, ship KV scales -- and a mechanistic account of why the recurrent half of a hybrid LLM is the easy half to quantize. Checkpoint: https://huggingface.co/minima-ai/mnma_qwen3.8_27b_nvfp4

Related papers

• On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability [57.03393882933515]
  We describe the architecture and ablations of Qwen3.8-Flash-Next, a sparse mixture-of-experts model with 125B parameters, 6B activated per token, and 51B parameters of n-gram embedding tables held off the accelerator.<n>On fourteen pre-training benchmarks the model leads the 397B-A17B predecessor on eight and trails it on the rest by at most 2.6 points, at 1/3 the activated parameters, 1/3 the training tokens, and roughly 1/9 the training FLOPs.
  arXiv  Detail & Related papers  (2026-08-31T06:35:07Z)
• RDQ: Residual Distribution Quantization for Large Language Models [3.7323766833175718]
  Post-training quantization of large language models degrades sharply below 4-bit precision.<n>We propose RDQ (Residual Distribution Quantization), a PTQ framework whose central contribution is Cascaded Error Compensation.<n>All output is standard group-128 asymmetric quantization, deployable on Qualcomm AIMET, GGUF, and any standard inference stack at zero runtime overhead.
  arXiv  Detail & Related papers  (2026-07-11T05:54:00Z)
• 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)
• MosaicQuant: Inlier-Outlier Disaggregation for Unified 4-Bit LLM Quantization [27.433770037419624]
  We propose a unified 4-bit quantization paradigm for large language models (LLMs)<n>MosaicQuant quantizes the full weight matrix into a dense 4-bit base component, where inliers are captured faithfully while outliers are inevitably quantized.<n>Experiments on LLaMA3 and Qwen3 demonstrate that MosaicQuant preserves near-FP16 accuracy while achieving up to $1.24times$ speedup over the W16A16 baseline.
  arXiv  Detail & Related papers  (2026-06-14T07:43:43Z)
• When Flat Minima Fail: Characterizing INT4 Quantization Collapse After FP32 Convergence [0.0]
  Post-training quantization assumes that a well-converged model is a quantization-ready model.<n>We show this assumption fails in a structured, measurable, and previously uncharacterized way.
  arXiv  Detail & Related papers  (2026-04-16T15:46:36Z)
• BWTA: Accurate and Efficient Binarized Transformer by Algorithm-Hardware Co-design [71.97035034203275]
  We analyze zero-point distortion in binarization and propose a Binary Weights & Ternary Activations (BWTA) quantization scheme.<n>For training, we propose Smooth Multi-Stage Quantization, combining a Levelwise Degradation Strategy and a MagnitudeAlignment Projection Factor.<n>Experiments show that BWTA approaches full-precision performance for Transformer-based models, with an average 3.5% drop on GLUE and less than 2% drop on five tasks.
  arXiv  Detail & Related papers  (2026-04-05T04:25:07Z)
• BPDQ: Bit-Plane Decomposition Quantization on a Variable Grid for Large Language Models [56.504879072674015]
  We propose Bit-Plane Decomposition Quantization (BPDQ), which constructs a variable quantization grid via bit-planes and scalar coefficients.<n>BPDQ enables serving Qwen2.5-72B on a single GTX 3090 with 83.85% GSM8K accuracy (vs. 90.83% at 16-bit)
  arXiv  Detail & Related papers  (2026-02-04T02:54:37Z)
• ResQ: Mixed-Precision Quantization of Large Language Models with Low-Rank Residuals [10.860081994662645]
  Post-training quantization of large language models (LLMs) holds the promise in reducing the prohibitive computational cost at inference time.<n>We propose ResQ, a PTQ method that pushes further the state-of-the-art.<n>We demonstrate that ResQ outperforms recent uniform and mixed precision PTQ methods on a variety of benchmarks.
  arXiv  Detail & Related papers  (2024-12-18T22:01:55Z)
• 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)
• QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving [52.31791050376249]
  Quantization can accelerate large language model (LLM) inference.<n>We introduce QoQ, a W4A8KV4 quantization algorithm with 4-bit weight, 8-bit activation, and 4-bit KV cache.<n>QServe improves the maximum achievable serving of Llama-3-8B by 1.2x on A100, 1.4x on L40S; and Qwen-721.5B by 2.4x on A100, 3.5x on L40S.
  arXiv  Detail & Related papers  (2024-05-07T17:59:30Z)
• HAWQV3: Dyadic Neural Network Quantization [73.11579145354801]
  Current low-precision quantization algorithms often have the hidden cost of conversion back and forth from floating point to quantized integer values. We present HAWQV3, a novel mixed-precision integer-only quantization framework.
  arXiv  Detail & Related papers  (2020-11-20T23:51:43Z)

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.