# TensorDex: A Compact, Tensor-Centric Storage System for Modern AI Models

arXiv: https://arxiv.org/abs/2604.17104 · HTML: https://mlsanity.com/paper/2604.17104/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 25 of its 89 references that are in the map.
- First posted on arXiv 18 Apr 2026.

## Builds on (most cited first, 25 of 25)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers](https://mlsanity.com/paper/2210.17323/) (2022)
- [Gemma 3 Technical Report](https://mlsanity.com/paper/2503.19786/) (2025)
- [SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models](https://mlsanity.com/paper/2211.10438/) (2022)
- [AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration](https://mlsanity.com/paper/2306.00978/) (2023)
- [Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs](https://mlsanity.com/paper/1603.09320/) (2016)
- [ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers](https://mlsanity.com/paper/2206.01861/) (2022)
- [SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression](https://mlsanity.com/paper/2306.03078/) (2023)
- [Q8BERT: Quantized 8Bit BERT](https://mlsanity.com/paper/1910.06188/) (2019)
- [Fast Approximate Nearest Neighbor Search With The Navigating Spreading-out Graph](https://mlsanity.com/paper/1707.00143/) (2017)
- [Model Stock: All we need is just a few fine-tuned models](https://mlsanity.com/paper/2403.19522/) (2024)
- [Knowledge is a Region in Weight Space for Fine-tuned Language Models](https://mlsanity.com/paper/2302.04863/) (2023)
- [Optimal Data-Dependent Hashing for Approximate Near Neighbors](https://mlsanity.com/paper/1501.01062/) (2015)
- [ByteCheckpoint: A Unified Checkpointing System for Large Foundation Model Development](https://mlsanity.com/paper/2407.20143/) (2024)
- [DeltaZip: Efficient Serving of Multiple Full-Model-Tuned LLMs](https://mlsanity.com/paper/2312.05215/) (2023)
- [ZipNN: Lossless Compression for AI Models](https://mlsanity.com/paper/2411.05239/) (2024)
- [Anatomy of a Machine Learning Ecosystem: 2 Million Models on Hugging Face](https://mlsanity.com/paper/2508.06811/) (2025)
- [An Empirical Analysis of Machine Learning Model and Dataset Documentation, Supply Chain, and Licensing Challenges on Hugging Face](https://mlsanity.com/paper/2502.04484/) (2025)
- [Dynamic Facility Location via Exponential Clocks](https://mlsanity.com/paper/1411.4476/) (2014)
- [QStore: Quantization-Aware Compressed Model Storage](https://mlsanity.com/paper/2505.04081/) (2025)
- [ZipLLM: Efficient LLM Storage via Model-Aware Synergistic Data Deduplication and Compression](https://mlsanity.com/paper/2505.06252/) (2025)
- [Forecasting Open-Weight AI Model Growth on HuggingFace](https://mlsanity.com/paper/2502.15987/) (2025)
- [NeurStore: Efficient In-database Deep Learning Model Management System](https://mlsanity.com/paper/2509.03228/) (2025)

## Abstract (arXiv)

Modern model hubs store hundreds of petabytes of large language models (LLMs), with fine-tuned variants dominating the storage footprint. These variants contain substantial cross-model redundancy that delta compression can exploit by storing only the difference between a target and a reference model. However, compression effectiveness depends critically on choosing a similar reference. At model-hub scale, this is challenging because model lineage metadata is often missing or unreliable, and different tensors within the same model may be most similar to tensors from different models. Consequently, model-level pairing leaves substantial redundancy unexploited. We present TensorDex, a lossless, tensor-centric storage system that performs delta compression at tensor granularity. TensorDex decomposes models into tensors, uses compact bit-level fingerprints to predict tensor-pair compressibility, and incrementally organizes tensors into multi-center clusters to select effective bases as new models arrive. Evaluated on 2,890 randomly sampled Hugging Face models, TensorDex reduces storage footprint by 70.5%, 37% lower than the state-of-the-art design. It achieves 22.9 GB/s compression and 28.4 GB/s decompression throughput, 3.86x and 1.49x faster than the next-best system, respectively.