When Your Fine-Tune Isn't Done Learning Yet
from My Weird Prompts
Your fine-tune's loss curve is dropping beautifully. So why is it still picking the wrong citation, drifting back into base-model habits halfway through an answer, or transcribing the same words wrong as the stock model? This episode works through what undertraining actually looks like in fine-tuned models — Whisper, LLMs, and speech models adapted to a single voice — and how to tell unfinished learning apart from bad data, a broken learning rate, or a capability you accidentally trained away. Along the way: why teacher forcing makes training loss a poor predictor of generated transcripts, why the new behavior showing up in six of ten samples is a classic undertraining signature, and why "just train longer" is sometimes the most expensive wrong answer available. Episode 6244 — myweirdprompts.com/n/6244 (permanent short link: myweirdprompts.com/665267) Listen on Android: MWP FM Player — myweirdprompts.com/apk
Embed this episode
Chapters
No transcript yet.