Sparse Upcycling: Inference Inefficient Finetuning

TL;DR

This work compares the effectiveness of sparse upcycling against continued pretraining (CPT) across different model sizes, compute budgets, and pretraining durations, and shows that sparse upcycling can achieve better quality, but comes with a significant inference cost, leading to 40% slowdowns in high-demand inference settings for larger models.

Abstract

Small, highly trained, open-source large language models are widely used due to their inference efficiency, but further improving their quality remains a challenge. Sparse upcycling is a promising approach that transforms a pretrained dense model into a Mixture-of-Experts (MoE) architecture, increasing the model's parameter count and quality. In this work, we compare the effectiveness of sparse upcycling against continued pretraining (CPT) across different model sizes, compute budgets, and pretraining durations. Our experiments show that sparse upcycling can achieve better quality, with improvements of over 20% relative to CPT in certain scenarios. However, this comes with a significant inference cost, leading to 40% slowdowns in high-demand inference settings for larger models. Our findings highlight the trade-off between model quality and inference efficiency, offering insights for practitioners seeking to balance model quality and deployment constraints.

Figures (4)

• Figure 1: Results from CPT vs. upcycling. The top row shows the relative improvements on Gauntlet Core Average as a function of additional training. The bottom row compares the final cross entropy scores of the upcycled and CPT models.
• Figure 2: Inference Speed of CPT vs. Upcycled Models.
• Figure 3: Latency vs. Throughput - 8B vs 47B MoE
• Figure 4: Absolute Gauntlet Scores for Dense vs. Upcycled Models.