SACQ: Structure-Aware Codebook Quantization for Ultra-Low-Bit Vision-Language-Action Models
Overview
This paper introduces SACQ, a weight-only vector-codebook quantization framework designed for Vision-Language-Action (VLA) models to enable deployment on resource-constrained robotic platforms. The method employs structure-aware codebook fitting, action-aware mixed-bit allocation, and two-stage post-quantization refinement. Evaluated on the GR00T N1.7 model, SACQ achieves 93.85% average success across four LIBERO suites at 1.5 bits per language/action weight, reducing memory footprint by 4.83x. It also retains 93.55% success with INT8 activation quantization while running 1.51x faster than BF16. On the SimplER benchmark, it achieves 63.76% and 52.27% success under Visual Matching and Variant Aggregation at 2 bits per weight, outperforming BF16 baselines.
Generated from attributed reports · Updated 2 hours ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · memory reduction factor: 4.83 other · Basis not reported
“el memory by 4.83x. With INT8 activation quantization and low-precision attention, it retains 93.55% success while running 1.51x faster than BF16. On SimplER, SACQ further achieves 63.76% and 52.27% success under Visual Matching and Variant Aggregation at 2 bits per language/action weight, compared to 58.90% and 53.45% for BF16, respectively.”
Exact source · revision 1Source owner not reported
Reported quantity · speedup factor: 1.51 other · Basis not reported
“el memory by 4.83x. With INT8 activation quantization and low-precision attention, it retains 93.55% success while running 1.51x faster than BF16. On SimplER, SACQ further achieves 63.76% and 52.27% success under Visual Matching and Variant Aggregation at 2 bits per language/action weight, compared to 58.90% and 53.45% for BF16, respectively.”
Exact source · revision 1Source owner not reported
Reported quantity · bits per language/action weight: 1.5 other · Basis not reported · Differing source assertions
“allocation based on the increase in final-action error as individual layers are compressed; and (3) two-stage post-quantization refinement that reassigns codeword indices with second-order error compensation and calibrates codebooks and row-group scales against the decoded action. Using GR00T N1.7, SACQ achieves 93.85% average success across four LIBERO suites at 1.5 bits per language/action weight while reducing mod”
Exact source · revision 1Source owner not reported
Reported quantity · bits per language/action weight: 2 other · Basis not reported · Differing source assertions
“el memory by 4.83x. With INT8 activation quantization and low-precision attention, it retains 93.55% success while running 1.51x faster than BF16. On SimplER, SACQ further achieves 63.76% and 52.27% success under Visual Matching and Variant Aggregation at 2 bits per language/action weight, compared to 58.90% and 53.45% for BF16, respectively.”
Exact source · revision 1Source owner not reported
Report timeline
Follow attributed reports and material updates.
- arXiv Robotics — research abstractsSignalSACQ: Structure-Aware Codebook Quantization for Ultra-Low-Bit Vision-Language-Action Models
This paper introduces SACQ, a weight-only vector-codebook quantization framework designed for Vision-Language-Action (VLA) models to enable deployment on resource-constrained robotic platforms. The method employs structure-aware codebook fitting, action-aware mixed-bit allocation, and two-stage post-quantization refinement. Evaluated on the GR00T N1.7 model, SACQ achieves 93.85% average success across four LIBERO suites at 1.5 bits per language/action weight, reducing memory footprint by 4.83x. It also retains 93.55% success with INT8 activation quantization while running 1.51x faster than BF16. On the SimplER benchmark, it achieves 63.76% and 52.27% success under Visual Matching and Variant Aggregation at 2 bits per weight, outperforming BF16 baselines.
Event coverage history
There is not enough continuous observation data to show a trend.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).