Remote KV caching for long-context LLM serving can severely degrade latency unless the store is sharded by parallel rank, ...
OneWM unifies causal training of world and action modeling, showing that learning action-conditioned dynamics improves ...
Velox accelerates long video rollouts by routing attention via geometric frustum slabs, reducing latency while preserving ...
Join the discussion on this paper page ...
Muse Spark Open model. A 816B parameter model with 42B Active parameters. BF16 and NVFP4 Weights.
Timestral is a family of time series foundation models for zero-shot forecasting, available in 5M, 28M, 123M, and 209M parameter versions. This repository contains the 209M checkpoint. Timestral ...
Timestral is a family of time series foundation models for zero-shot forecasting, available in 5M, 28M, 123M, and 209M parameter versions. This repository contains the 123M checkpoint. For more ...
The method converts pre-trained transformers into elastic models that adapt continuously to varying computational budgets by ...
Cortex turns context-induced performance gains into mergeable parameter updates, enabling LLMs to internalize new knowledge while preserving existing capabilities for continual le ...
Hi everyone, author here! ActionOPD distills agent decisions through semantic agreement with black-box teachers, without requiring teacher token probabilities or tokenizer alignme ...
Join the discussion on this paper page Player-1: Learning to Play from Synthetic Experience ...
Join the discussion on this paper page Are We Testing Too Much? A Survey of Benchmark Compression for Efficient Evaluation of Large Language Models ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results