Baobao Algorithm Notes
Oct 25, 2023 · Artificial Intelligence
How Mixed Data Shapes LLaMA SFT: Scaling Trends, Conflict Zones, and the DMT Remedy
This article investigates how mixing data from mathematical reasoning, code generation, and general instruction-following tasks influences supervised fine‑tuning of LLaMA models, revealing distinct scaling curves, resource‑dependent performance conflicts, and a two‑stage DMT strategy that mitigates catastrophic forgetting while boosting overall capability.
DMT StrategyLlamaModel Fine‑tuning
0 likes · 14 min read
