{{ message }}
Repository navigation
[Resource]: CacheMind: Trace-Grounded Reasoning for Cache Replacement #15
Description
Resource title
CacheMind: From Miss Rates to Why -- Natural-Language, Trace-Grounded Reasoning for Cache Replacement
URL
https://arxiv.org/abs/2602.12422
Resource type
Paper
Venue or source
ASPLOS 2026
Date or year
2026
Author(s)
Kaushal Mhapsekar; Azam Ghanbari; Bita Aslrousta; Samira Mirbagher-Ajorpaz
DOI
10.1145/3779212.3790136
Primary topic
Architecture 2.0
Short description
CacheMind uses retrieval-augmented generation over CPU cache traces to answer natural-language questions about cache behavior and replacement policies, linking architectural analysis to trace evidence. The paper introduces CacheMindBench for evaluating trace-grounded and policy-specific reasoning; the authors provide public code and benchmark questions.
Coauthor disclosure
I am Samira Mirbagher Ajorpaz, a coauthor of this paper.
Activity
github-actions commentedon Oct 4, 2026
Thanks for the submission. I parsed CacheMind: From Miss Rates to Why -- Natural-Language, Trace-Grounded Reasoning for Cache Replacement
and opened a pull request to add it to the reading list. A maintainer
will review and merge it, and it will appear on the site shortly after.