BaKlaVa -- Budgeted Allocation of KV cache for Long-context Inference
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909507691479040 |
|---|---|
| author | Gulhan, Ahmed Burak Chitty-Venkata, Krishna Teja Emani, Murali Kandemir, Mahmut Vishwanath, Venkatram |
| author_facet | Gulhan, Ahmed Burak Chitty-Venkata, Krishna Teja Emani, Murali Kandemir, Mahmut Vishwanath, Venkatram |
| contents | In Large Language Model (LLM) inference, Key-Value (KV) caches (KV-caches) are essential for reducing time complexity. However, they result in a linear increase in GPU memory as the context length grows. While recent work explores KV-cache eviction and compression policies to reduce memory usage, they often consider uniform KV-caches across all attention heads, leading to suboptimal performance. We introduce BaKlaVa, a method to allocate optimal memory for individual KV-caches across the model by estimating the importance of each KV-cache. Our empirical analysis demonstrates that not all KV-caches are equally critical for LLM performance. Using a one-time profiling approach, BaKlaVa assigns optimal memory budgets to each KV-cache. We evaluated our method on LLaMA-3-8B, and Qwen2.5-7B models, achieving up to a 70\% compression ratio while keeping baseline performance and delivering up to an order-of-magnitude accuracy improvement at higher compression levels. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_13176 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | BaKlaVa -- Budgeted Allocation of KV cache for Long-context Inference Gulhan, Ahmed Burak Chitty-Venkata, Krishna Teja Emani, Murali Kandemir, Mahmut Vishwanath, Venkatram Machine Learning Artificial Intelligence In Large Language Model (LLM) inference, Key-Value (KV) caches (KV-caches) are essential for reducing time complexity. However, they result in a linear increase in GPU memory as the context length grows. While recent work explores KV-cache eviction and compression policies to reduce memory usage, they often consider uniform KV-caches across all attention heads, leading to suboptimal performance. We introduce BaKlaVa, a method to allocate optimal memory for individual KV-caches across the model by estimating the importance of each KV-cache. Our empirical analysis demonstrates that not all KV-caches are equally critical for LLM performance. Using a one-time profiling approach, BaKlaVa assigns optimal memory budgets to each KV-cache. We evaluated our method on LLaMA-3-8B, and Qwen2.5-7B models, achieving up to a 70\% compression ratio while keeping baseline performance and delivering up to an order-of-magnitude accuracy improvement at higher compression levels. |
| title | BaKlaVa -- Budgeted Allocation of KV cache for Long-context Inference |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2502.13176 |