Research map / Architectures and efficiency / Quantization
Accelerating Large-Scale Inference with Anisotropic Vector Quantization
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use score-aware quantization loss instead of reconstruction loss | Product quantization · Amazon-670k, 256 bits | Recall 1@1: 0.652 -> 0.656 | Helped |
| Use score-aware quantization loss instead of reconstruction loss | Product quantization · Amazon-670k, 512 bits | Recall 1@1: 0.737 -> 0.744 | Helped |
| Use score-aware quantization loss instead of reconstruction loss | Product quantization · Amazon-670k, 1024 bits | Recall 1@1: 0.778 -> 0.812 | Helped |
| Use score-aware quantization loss instead of reconstruction loss | Product quantization · Amazon-670k, 2048 bits | Recall 1@1: 0.782 -> 0.875 | Helped |
| Replace reconstruction loss with score-aware quantization loss | SGH · SIFT1M, 64 bits | Recall 1@1: 0.028 -> 0.071 | Helped |
| Replace reconstruction loss with score-aware quantization loss | SGH · SIFT1M, 128 bits | Recall 1@1: 0.073 -> 0.196 | Helped |
| Replace reconstruction loss with score-aware quantization loss | SGH · SIFT1M, 256 bits | Recall 1@1: 0.142 -> 0.362 | Helped |
| Use score-aware loss rather than reconstruction loss | Product quantization · Glove1.2M | Recall 1@10 | Helped |
| Use score-aware loss rather than reconstruction loss | Product quantization · Glove1.2M | Top-1 inner-product relative error | Helped |
Closest papers by meaning
- Qinco2: Vector Compression and Search with Improved Implicit Neural Codebooks (ICLR 2025 · similarity 0.86)
- Structured Transforms for Low-Overhead Quantization of Language Models (NeurIPS 2026 · similarity 0.84)
- NSNQuant: A Double Normalization Approach for Calibration-Free Low-Bit Vector Quantization of KV Cache (NeurIPS 2025 · similarity 0.84)
- Residual Quantization with Implicit Neural Codebooks (ICML 2024 · similarity 0.84)
- MagR: Weight Magnitude Reduction for Enhancing Post-Training Quantization (NeurIPS 2024 · similarity 0.84)
- Extreme Compression of Large Language Models via Additive Quantization (ICML 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.