Towards enhanced LLM pretraining: Dynamic checkpoint merging via generation quality.

Zecheng Wang, Deyuan Liu, Chunshan Li, Dianhui Chu, Weipeng Chen, Bingning Wang, Dianbo Sui

21 Jan 2026 (modified: 21 Jan 2026)Inf. Fusion 2026EveryoneRevisionsCC BY-SA 4.0
Loading