[PDG 434] Scaling LLM Test-Time Compute (can be More Effective than Parameters)

Link to article: https://arxiv.org/pdf/2408.03314
Title: Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters
Track: Scaling Laws
Content: We study how effectively LLM performance improves when given additional inference-time compute. Analyzing two strategies—verifier-guided search and adaptive response refinement—we find effectiveness varies by prompt difficulty. A "compute-optimal" strategy improves efficiency by over 4× compared to best-of-N sampling, and smaller models with inference-time compute can outperform 14× larger models under equal FLOPs budgets.
Slack link: ml-ka.slack.com, channel: #pdg. Please join us -- if you cannot join, please message us here or to mlpaperdiscussiongroupka@gmail.com.

In the Paper Discussion Group (PDG) we discuss recent and fundamental papers in the area of machine learning on a weekly basis. If you are interested, please read the paper beforehand and join us for the discussion. If you have not fully understood the paper, you can still participate – everyone is welcome! You can join the discussion or simply listen in. The discussion is in German or English depending on the participants.