technique

W4A4 quantization

techniqueactiveprovisionalw4a4-quantization-7bc687bd·1 events·first seen 20d ago

Aliases: W4A4 quantization

Co-occurring entities

Pi 0.5 SVD-Hadamard rotation LIBERO Vision-Language-Action model GR00T N1.5 Linear Diffusion Transformer Omega-QVLA

More like this (12)

INT4 quantisation INT4 Quantization 1.58-bit quantization Channel-wise Vector Quantization Power-of-Two (PoT) Quantization Vector Quantization quantization-induced degradation binary quantization scalar quantization quantization quantization-aware training KV Cache Quantization

Recent events (1)

7arXiv · cs.LG·20d ago·source ↗

Ω-QVLA: Training-Free W4A4 Quantization for Full Vision-Language-Action Models Including Diffusion Action Heads

Omega-QVLA is a post-training quantization framework that compresses both the LLM backbone and the diffusion-based action head of VLA models to uniform W4A4 precision without mixed-precision schemes or fine-tuning. It combines composite SVD-Hadamard rotation for weight energy equalization with per-step DiT activation scaling to handle dynamic-range drift across denoising steps. On the LIBERO benchmark, it achieves 98.0% and 87.8% task success on Pi 0.5 and GR00T N1.5 respectively—matching or exceeding FP16 baselines—while reducing static memory footprint by 71.3%. Real-world manipulation experiments confirm the approach generalizes beyond simulation.

Inference Economics Agent and Tool Ecosystem Pi 0.5 SVD-Hadamard rotation LIBERO +6 more