model: | |
learning_rate: 0.0003 | |
target: celle_main.CELLE_trainer | |
params: | |
ckpt_path: model.ckpt | |
condition_model_path: | |
condition_config_path: nucleus_vqgan.yaml | |
vqgan_model_path: | |
vqgan_config_path: threshold_vqgan.yaml | |
image_key: threshold | |
num_images: 2 | |
dim: 480 | |
num_text_tokens: 33 | |
text_seq_len: 1000 | |
depth: 68 | |
heads: 16 | |
dim_head: 64 | |
attn_dropout: 0.1 | |
ff_dropout: 0.1 | |
attn_types: full | |
rotary_emb: true | |
fixed_embedding: true | |
text_embedding: esm2 | |
loss_img_weight: 1 | |
loss_cond_weight: 1 |