When AI Agrees Too Much: Manipulating Automated Code Review via Bias Injection [pdf]dimitro.gr·2 pts·matt_d·0
TorchTPU: Hybrid Eager/Compiled Architecture for Native PyTorch Execution on TPUarxiv.org·1 pts·matt_d·1
Reactive Optimizers: rewrite the optimizer's source code during trainingxingjianbai.com·32 pts·matt_d·3
Agency: Open-source platform for debugging, profiling, and analyzing AI agentsgithub.com·4 pts·matt_d·1
A New Lens on Multi-GPU Comms: API-Level Memory Models for Portable Optimizationarxiv.org·2 pts·matt_d·0
Architecting for HW Diversity: Meta's E2E Co-Design for AI Training on AMD GPUsarxiv.org·3 pts·matt_d·0
The Vermilion Experiment: Implementing a Lean Back End for Verusproofsandintuitions.net·2 pts·matt_d·0
Beyond Zero-Skipping: Redundancy-Aware Arch. and Systems for Efficient Inference [video]youtube.com·1 pts·matt_d·0
AbridgeKV: Compressing Long Reasoning into the KV Cache of Learned Pause Tokensgithub.com·1 pts·matt_d·0