dongchen06 / dmweeds Goto Github PK

The first step of my training was to train diffusion.
Here are my parameters：
MODEL_FLAGS="--image_size 256 --num_channels 256 --num_res_blocks 2 --learn_sigma True --use_scale_shift_norm true --attention_resolutions 32,16,8 --num_head_channels 64"
DIFFUSION_FLAGS="--resblock_updown True --diffusion_steps 1000 --noise_schedule linear --rescale_learned_sigmas False --rescale_timesteps False "
TRAIN_FLAGS="--lr 1e-4 --batch_size 2 --dropout 0.1"
python scripts/image_train.py --data_dir /tmp/pycharm_project_123/datasets/image $MODEL_FLAGS $DIFFUSION_FLAGS $TRAIN_FLAGS
If I continue to want to train guided_classifier
Do you need to change any codes?
thank you

Add the above MODEL_FALGS to class_cond True
then python classifier_train.py ........ ?

Hello, I am interested in your work.

I was wondering how to run this model?

Hello, when I tried to replicate your transfer diffusion experiment using OpenAI's open source guided diffusion, I don't know how to choose batch size and microbatch. There is definitely no way to use a batch size of 128 in the case of a single GPU, and your LR will also be reduced in the case of reducing the batch size?

How can I use your model to train my own data set Thank you