STAPLE: Static Sparse Upcycling of Pretrained Checkpoints via Embedding Modules

Key points

• Sparse computation can scale the parametric capacity of LLMs without proportionally increasing active computation, but pretraining sparse architectures remains expensive and data-hungry.1
• Existing sparse upcycling methods mainly target Mixture-of-Experts models, where expert homogeneity, sample inefficiency, and system overheads can limit practical scalability.1
• Static sparse memory offers predictable access and token-specific updates that can help address these challenges, but naively integrating it into pretrained computation yields poor returns.1
• Importantly, despite introducing billions of parameters late in training, our host-offloaded implementation keeps generation throughput within 1.5% and training throughput within 3% of the dense baseline, while caching only the 5% most frequent embeddings on GPU.1

Sources (1)

• [1]STAPLE: Static Sparse Upcycling of Pretrained Checkpoints via Embedding Modules
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 06:21 PM

  In this work, we introduce STAPLE, a recipe built upon STEM for integrating static sparse memory into pretrained models.

  Sparse computation can scale the parametric capacity of LLMs without proportionally increasing active computation, but pretraining sparse architectures remains expensive and data-hungry.

Extractive summary: sentences quoted from the sources.