You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This repository is a CacheLib prototype of TARDIS, a learned cache eviction policy as part of the LDOS Expeditions project at UT Austin. It implements LRUForgive and S3FIFOForgive with a log-structured multithreaded architecture, and is evaluated against LRU & S3FIFO on real traces to to measure miss-ratio improvement under high eviction pressure.
Neural network built from scratch in NumPy to classify data-center workloads based on thermal load indicators (CPU utilization, runtime), using Google cluster traces.
This repository contains a set-associative cache simulator in C with four eviction policies: LRU, LFU, Bélády's OPT, and an online perceptron-based policy that scores by recency, frequency, and per-PC weights updated from a circular eviction history to explore whether a lightweight ML model can approximate OPT at runtime with no offline training.
Comparative analysis of Mamba vs. Transformers trained from scratch. Benchmarking Mamba's linear O(N) scaling and constant-time inference against quadratic attention mechanisms.