Open sorobedio opened 11 months ago
Hi, we apologize for the confusion. When training a vae, we allow for either using:
--precompute_checkpoint
, to use, as training data for the VAE, samples from a trained HNET--few_shot_checkpoint
, to use, as training data for the VAE, checkpoints from the train_few_shot.py
script, that simply trains few shot simple networks.So, you can either:
train_few_shot.py
, use the output to train an HNET with precompute_adaptation.py
, and use this HNET to generate data for the VAE with train_vae.py --precompute_checkpoint [...]
train_few_shot.py
, and directly use the output dataset ot train a VAE with train_vae.py --few_shot_checkpoint [...]
thank you for the prompt reaction.
what i did was.
1- python scripts/train_few_shot.py [...] perfectly executed with no problem
My confusion was in the argument parse configuration parameters list:
for example with precompute_adaptation.py
we have the following required parameters
parser.add_argument('--few_shot_checkpoint', type=str, help='required') parser.add_argument('--vae_checkpoint', type=str, help='required')
and with train_vae
we have
parser.add_argument('--few_shot_checkpoint', type=str, help='required') parser.add_argument('--precompute_checkpoint', type=str, help='required')
Therefore, precompute_adaptation.py
requires checkpoint from the run of train_vae.py
which requires the checkpoint of precompute_adaptation.
Thank you. the training is going well for the vae. i will close the issue after finishing the vae training.
VAE is normanly trained after: python scripts/precompute_adaptation.py (--few_shot_checkpoint | --vae_checkpoint ) [...]
But precompute adaptation requires vae_checkpoint and vae training requires precompute_checkpoint I am a little confused. can you provide the exact training command ?