# Meta-Soft: Leveraging Composable Meta-Tokens for Context-Preserving KV Cache Compression

arXiv: https://arxiv.org/abs/2605.22337 · HTML: https://mlsanity.com/paper/2605.22337/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving.
- Builds on 24 of its 28 references that are in the map.
- First posted on arXiv 21 May 2026.

## Placed in

- Efficient LLM inference and serving / [KV cache optimization](https://mlsanity.com/inference/kv-cache-optimization/)

## Builds on (most cited first, 24 of 24)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [The Pile: An 800GB Dataset of Diverse Text for Language Modeling](https://mlsanity.com/paper/2101.00027/) (2020)
- [Efficient Streaming Language Models with Attention Sinks](https://mlsanity.com/paper/2309.17453/) (2023)
- [LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding](https://mlsanity.com/paper/2308.14508/) (2023)
- [RULER: What's the Real Context Size of Your Long-Context Language Models?](https://mlsanity.com/paper/2404.06654/) (2024)
- [Token Merging: Your ViT But Faster](https://mlsanity.com/paper/2210.09461/) (2022)
- [SnapKV: LLM Knows What You are Looking for Before Generation](https://mlsanity.com/paper/2404.14469/) (2024)
- [Compressive Transformers for Long-Range Sequence Modelling](https://mlsanity.com/paper/1911.05507/) (2019)
- [Efficiently Scaling Transformer Inference](https://mlsanity.com/paper/2211.05102/) (2022)
- [Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs](https://mlsanity.com/paper/2310.01801/) (2023)
- [Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Time](https://mlsanity.com/paper/2305.17118/) (2023)
- [PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling](https://mlsanity.com/paper/2406.02069/) (2024)
- [Learning to Compress Prompts with Gist Tokens](https://mlsanity.com/paper/2304.08467/) (2023)
- [Adapting Language Models to Compress Contexts](https://mlsanity.com/paper/2305.14788/) (2023)
- [No Token Left Behind: Reliable KV Cache Compression via Importance-Aware Mixed Precision Quantization](https://mlsanity.com/paper/2402.18096/) (2024)
- [ZipCache: Accurate and Efficient KV Cache Quantization with Salient Token Identification](https://mlsanity.com/paper/2405.14256/) (2024)
- [On the Efficacy of Eviction Policy for Key-Value Constrained Generative Language Model Inference](https://mlsanity.com/paper/2402.06262/) (2024)
- [D2O: Dynamic Discriminative Operations for Efficient Long-Context Inference of Large Language Models](https://mlsanity.com/paper/2406.13035/) (2024)
- [LoCoCo: Dropping In Convolutions for Long Context Compression](https://mlsanity.com/paper/2406.05317/) (2024)
- [Lookahead Q-Cache: Achieving More Consistent KV Cache Eviction via Pseudo Query](https://mlsanity.com/paper/2505.20334/) (2025)
- [ZSMerge: Zero-Shot KV Cache Compression for Memory-Efficient Long-Context LLMs](https://mlsanity.com/paper/2503.10714/) (2025)
- [Judge Q: Trainable Queries for Optimized Information Retention in KV Cache Eviction](https://mlsanity.com/paper/2509.10798/) (2025)

## Abstract (arXiv)

The KV cache used in large language models has linearly growing time complexity, so LLMs face memory blow-up and reduced decoding efficiency when they process long contexts. Current KV Cache eviction has become an important research direction; however, existing methods based on fixed Soft Tokens (e.g., Judge Q) rely on a static parameter set as the query to evaluate the importance of KV pairs, so they cannot adapt dynamically to different input prompts, and they cannot precisely capture complex and changing task relevance. Also, evicted KV pairs are discarded permanently, so this causes irreversible information loss and context breaks. To address this problem, we propose Meta-Soft, a dynamic compression framework based on probe-driven context integration. Specifically, we build a meta-library with a learnable orthogonal basis matrix L, and we use a selector network with Gumbel-Softmax to produce differentiable sparse combination weights, so we dynamically synthesize the most targeted k Soft Tokens from the input prompt features. We append these Soft Tokens to the end of the input sequence to probe key information. We also introduce an attention-flow based integration mechanism, which redistributes the semantic information of removed tokens into retained tokens, and this keeps the dropped context information effectively. Experiments on multiple datasets show that our method outperforms existing state-of-the-art eviction methods and provides a new solution for KV Cache compression.