Tag
Qwenputer：在 Qwen3.5-4B 权重内部嵌入完整 RISC-V CPU，利用模型自身的 Gated DeltaNet 层充当内存，可在前向传播中直接运行 C 和 MicroPython，无需外部运行时。在 DeepMind 数学基准上达到 97.5–100% 的成绩。
nesformer 是一个开源项目，通过编译器将 NES 的 6502 CPU、2KB RAM 与卡带 ROM 直接编译进一个未经训练的 Qwen3.5 模型权重中，让模型逐 token 预测 CPU 执行日志来运行《超级马力欧》等游戏，且帧输出与传统模拟器完全一致。
Qwenputer is an experimental Qwen 3.5 4B model with a fully functioning, deterministic RISC-V CPU compiled into its weights, using the model's own Gated DeltaNet layers as exact RAM to run arithmetic, C and MicroPython entirely within forward passes — no tools or external runtime needed.
HLA-WM提出一种训练免费的混合线性注意力框架，通过几何引导的检索与递归线性状态重组合来缓解GatedDeltaNet在长时视频世界模型中的远距离遗忘问题，在60秒基准上以12倍内存节省同时提升所有一致性与相机控制指标。
TinyCeNN-LM proposes a quality-gated post-training conversion framework using CeNN-inspired cellular-recurrent layers to replace attention in pretrained language models, balancing efficiency with performance preservation.
SpectralShift introduces a spectral reparameterization approach to effectively extend the context window of Gated DeltaNet models by reshaping the decay spectrum, improving long-context capabilities through continual pretraining.
This paper explains why fully quantizing the recurrent Gated DeltaNet layers in hybrid LLMs to 4-bit NVFP4 preserves accuracy across long-context benchmarks, detailing mechanisms like outlier localization and robust delta-rule dynamics.
Apple's Lily engine optimizes on-device LLM inference for Apple silicon by leveraging unified memory and hardware, outperforming MLX-LM, and is tuned for the Qwen3.6-35B-A3B model's architecture.
This paper introduces MixerLoop, a method that allocates recurrent compute by selectively looping the mixer component in language models while applying the feed-forward network once, achieving performance improvements with reduced computational costs.
This paper compares xLSTM, Mamba-2, and Gated DeltaNet on complex sequence modeling tasks and finds xLSTM superior due to its enhanced state tracking and memory dynamics, validated on synthetic length-generalization tasks.