search

yanghb22-fdu / Hi3D-Official

[MM24] Official codes and datasets for ACM MM24 paper "Hi3D: Pursuing High-Resolution Image-to-3D Generation with Video Diffusion Models".
MIT License
215 stars 12 forks source link

About finetune model #21

Closed zbtzbt44 closed 6 days ago

zbtzbt44 commented 1 week ago

Thanks for your excellent eork!I have a problem with finetuning the model("first stage.pt").When I train with the vanilla SVD model,I can get the following 2 model weights."mp_rank_00_model_states.pt" has the same size as "first stage.pt",but I cannot get the "zero_pp_rank_0_mp_rank_00_optim_states.pt" matching the "first stage.pt". Therefore,if I want to finetune the model with more datasets,I replace "first stage.pt" with "mp_rank_00_model_states.pt",but how to get the "zero_pp_rank_0_mp_rank_00_optim_states.pt"? image

zbtzbt44 commented 6 days ago

I solve it.Just the first weight is OK.

2hiTee commented 17 hours ago

I solve it.Just the first weight is OK.

Hi, so if I fine-tuned the model, which one should I use for inference? or continue fituning? Thank you!