How to use save weights

[vikas784]

So actually i am trying to train the model on my own data and i have own finetuned sdxk model , so i m confused on two things ,

First one is for step aware preference model i got two weights like this

pytorch_model-00001-of-00002.bin
pytorch_model-00002-of-00002.bin
pytorch_model.bin.index.json

so how should i combine it to make it one .pt file as u have used in the SPO training code for which the name is model_ckpts/sdxl_step-aware_preference_model.bin and

after that when u used this for training the SPO model with 4k prompts

i got 10 checkpoints folder and one lora.safetensor so how i have to use it to make ne safetensors for sdxl as u havent gave information for it

[RockeyCoss]

1. It looks like your weights are in the Hugging Face format. To convert them to a .pt file, first load them with from_pretrained(), then save the model’s state dictionary with:

torch.save(model.state_dict(), 'name.pt')

2. To generate the safetensors, first load the weights into your model (see this snippet), then export the LoRA weights in safetensor format (see this snippet).