Question about CFG used in CogVideoX-5B-I2V
Author: johnren-codeCreated Oct 13, 2025Updated Oct 13, 2025
When I finetune CogVideoX-5B-I2V in SFT, I find it use class free guidance for text condition in inference phase(default 5), but there is no text condition dropout during training phase in finetune code, may I ask is it a normal behaviour?
Source: zai-org/CogVideo