Pre-training LLM without Learning Rate Decay Enhances Supervised Fine-Tuning

TL;DR

Pre-training large language models with a constant learning rate after warmup (WSO) improves their adaptability and performance in downstream tasks compared to decay-based learning rate schedulers.

Contribution

This study demonstrates that avoiding learning rate decay during pre-training leads to models with flatter minima and better downstream performance, challenging conventional training practices.

Findings

WSO outperforms decay-based schedulers after supervised fine-tuning

Models trained with WSO have flatter loss landscape minima

Avoiding LR decay enhances downstream task adaptability

Abstract

We investigate the role of learning rate scheduling in the large-scale pre-training of large language models, focusing on its influence on downstream performance after supervised fine-tuning (SFT). Decay-based learning rate schedulers are widely used to minimize pre-training loss. However, despite their widespread use, how these schedulers affect performance after SFT remains underexplored. In this paper, we examine Warmup-Stable-Only (WSO), which maintains a constant learning rate after warmup without any decay. Through experiments with 1B and 8B parameter models, we show that WSO consistently outperforms decay-based schedulers in terms of performance after SFT, even though decay-based schedulers may exhibit better performance after pre-training. The result also holds across different regimes with mid-training and over-training. Loss landscape analysis further reveals that decay-based schedulers lead models into sharper minima, whereas WSO preserves flatter minima that support adaptability. These findings indicate that applying LR decay to improve pre-training metrics may compromise downstream adaptability. Our work also provides practical guidance for training and model release strategies, highlighting that pre-training models with WSO enhances their adaptability for downstream tasks.