2 papers
cs.LG2026
RaMP: Runtime-Aware Megakernel Polymorphism for Mixture-of-Experts
Vyom Sharma, Debajyoti Datta
The optimal kernel configuration for Mixture-of-Experts (MoE) inference depends on both batch size and the expert routing distribution, yet production systems dispatch from batch s…
cs.LG2026
ManifoldKV: Training-Free KV Cache Compression via Euclidean Outlier Detection
Debajyoti Datta, Trishala Neeraj, Bibek Paudel +2
Long-context inference is constrained by KV-cache memory, which grows linearly with sequence length; KV-cache compression therefore hinges on reliably selecting which past tokens t…