search

ai-computing / aicomp

Other
6 stars 0 forks source link

Memory leak in gptj 6B model training #13

Open ememos opened 3 months ago

ememos commented 3 months ago

Memory leak occurs when running the GPT-J 6B model.

ememos commented 2 months ago

The process of storing the results of the loss calculation in the data structure has been fixed.