hard20 pts · first answer countsfine-tuningmixed-precision
NaN loss in fp16 fine-tuning
Full fine-tuning in fp16 on an A100 hits NaN loss after a few hundred steps, even with a loss scaler. Which change most often fixes it on this hardware?
Discussion 0
Join free to post in the discussion.
Related questions
- What full fine-tuning changesMachine Learning Engineer
- Why begin from a pretrained modelMachine Learning Engineer
- Training only a new headMachine Learning Engineer
- Learning rate for fine-tuningMachine Learning Engineer
- Shape of SFT training dataMachine Learning Engineer
- Reading train and validation lossMachine Learning Engineer
- Meaning of PEFTMachine Learning Engineer
- What the LoRA rank r setsMachine Learning Engineer