T3C实现测试时按需压缩，三重保障模型效率与可靠性。
T3C: Test-Time Tensor Compression with Consistency Guarantees
- 测试时动态调节张量秩与精度，自动匹配硬件预算
- ResNet-50延迟1.18ms（38MB），ViT-B/16延迟2.30ms（59MB）
- 提供可验证的稳定性保证，适合边缘部署场景
我们提出T3C，一种训练一次、测试时按预算动态压缩的框架，将张量秩和精度作为可控部署参数。T3C结合弹性张量分解（最大秩可调）与秩相关混合精度量化，并通过轻量级控制器将延迟/能耗/大小预算映射为各层的秩/位宽分配；策略可对齐硬件特性且随预算单调变化。利用谱代理与激活统计量快速计算逐层一致性证书，上界约束输出偏移，训练正则化，仅需极低开销。在ImageNet-1k上，T3C推动视觉模型性能边界：ResNet-50在准确率下降≤0.5%条件下，中位延迟达1.18ms（模型38MB），优于PTQ-8b（1.44ms，88MB）；ViT-B/16实现2.30ms中位延迟（59MB），超越强基线。单一T3C检查点即可按需提供可预测、带证书保障的精度-延迟-尺寸权衡。
原文摘要 · Abstract (English)
We present T3C, a train-once, test-time budget-conditioned compression framework that exposes rank and precision as a controllable deployment knob. T3C combines elastic tensor factorization (maintained up to a maximal rank) with rank-tied mixed-precision quantization and a lightweight controller that maps a latency/energy/size budget token to per-layer rank/bit assignments; the policy snaps to hardware-aligned profiles and is monotone in the budget. A fast, layerwise consistency certificate, computed from spectral proxies and activation statistics, upper-bounds logit drift and regularizes training, yielding a practical reliability signal with negligible overhead. On ImageNet-1k, T3C shifts the vision Pareto frontier: for ResNet-50 at matched accuracy (\leq 0.5% drop), p50 latency is 1.18ms with a 38MB model, outperforming PTQ-8b (1.44ms, 88MB); for ViT-B/16, T3C reaches 2.30ms p50 with 59MB, improving over strong PTQ/QAT baselines. A single T3C checkpoint therefore provides predictable, certificate-backed accuracy-latency-size trade-offs on demand across devices.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。