FoldQuantVLA：视觉-语言-动作模型的原生低比特量化框架
FoldQuantVLA: Native Low-Bit Quantization of Vision-Language-Action Models via Consistent Folding
想在 Jetson 上把机器人 VLA 模型压到 4-bit 还不掉成功率？GR00T 和 π0.5 的实测数据都有。
论文提出 FoldQuantVLA，一个面向视觉-语言-动作（VLA）模型的训练后量化框架，无需重训策略即可完成 W4A4 量化。它把通道缩放与分块 Hadamard 变换结合动态 per-token 量化，通过定制 TensorRT 插件在语言主干和动作专家中执行原生整数推理，支持 Ada GPU 和 Jetson AGX Orin。在 LIBERO、SimplerEnv 和两个真实机器人平台上，三个 GR00T 检查点与 π0.5 的 W4A4 推理相比浮点 TensorRT 在 Orin 上提速 1.20 到 1.33 倍，桌面端提速 1.25 到 1.52 倍。将语言注意力输出和前馈下投影保留为 W8A8 后，GR00T N1.7 在四个真实机器人任务上的成功率从 80.0% 提到 92.5%（每配置 80 次试验），Orin 上额外延迟仅 1 ms。
FoldQuantVLA: Native Low-Bit Quantization of Vision-Language-Action Models via Consistent Folding
Low-bit vision-language-action inference must reduce observation-to-action latency while preserving robot behavior. We present FoldQuantVLA, a post-training quantization framework that carries a consistent activation representation through calibration, weight rounding, and native integer execution. It combines channel scaling and block Hadamard transforms with dynamic per-token quantization, without policy retraining. Custom TensorRT plugins execute projections in both the language backbone and iterative action expert with four-bit weights and activations (W4A4) on Ada GPUs and Jetson AGX Orin. Evaluation spans LIBERO, SimplerEnv, and two robot platforms. Across three GR00T checkpoints and $π_{0.5}$, W4A4 achieves $1.20$ to $1.33\times$ speedups over floating-point TensorRT on Orin and $1.25$ to $1.52\times$ on desktop. Retaining language attention-output and feed-forward down projections at eight bits (W8A8) improves held-out action fidelity on all four checkpoints. Across four real-robot tasks, this configuration raises observed GR00T N1.7 success from $80.0\%$ with uniform W4A4 to $92.5\%$ over 80 trials per configuration, with a measured additional Orin latency of 1 ms.