Modern High-Performance Computing workflows rely on Checkpoint/Restart (C/R) mechanisms for execution continuity, but their inherent task and data heterogeneity demands adaptability from the underlying C/R solution. In a case study with Montage, C/R with static configurations leads to an I/O overhead of $1.6 \times$ an...
Olga Kogiou, Hariharan Devarajan, Chen Wang et al.· IEEE International Conferenc...· 0 citations
This work conducts an empirical study of three state-of-the-art KV cache management frameworks: vLLM, InfiniGen, and H2O, revealing the most suitable selection and configuration of KV cache strategies under memory and performance constraints.
Oteo Mamo, Olga Kogiou, Hyunji Yi et al.· arXiv.org· 1 citation
SALT, a model-agnostic extractive framework that organizes per-sentence keywords into a trie ordered by sentence frequency (SF), a lightweight, reusable proxy for document thematic structure, reduces the prefill computation and memory cost of long-context prompts while remaining composable with KV-cache methods that ta...
