This paper proposes a new decoding strategy for looped language models, called late entry, which allows the model to skip initial passes and directly use a deep context cache, resulting in faster decoding times. Practitioners might care about this approach because it could lead to significant speedups in applications where decoding speed is crucial.
Firehose
Filtered to tagged “sequential decoding” · clear filters
Browse: People · Companies · Papers · Podcasts · Hacker News · Deep dives
Browse by tag
artificial intelligence 63open-weight models 19agentic coding 17continual learning 16AI 15AI agents 11reinforcement learning 9datasette 8AI safety 7cybersecurity 7Databricks 7natural language processing 7AI Gateway 6large language models 6Reinforcement learning 6vision-language models 6computer vision 5Diffusion models 5language models 5machine learning 5open-source 5Agentic AI 4AI ethics 4benchmarks 4education 4productivity 4recursive self-improvement 4Recursive self-improvement 4space 4tech 4