| { | |
| "experiment": { | |
| "tokenizer_checkpoint": "tatitok_bl32_vae.bin", | |
| "generator_checkpoint": "maskgen_kl_l.bin" | |
| }, | |
| "model": { | |
| "vq_model": { | |
| "quantize_mode": "vae", | |
| "token_size": 16, | |
| "vit_enc_model_size": "base", | |
| "vit_dec_model_size": "large", | |
| "vit_enc_patch_size": 16, | |
| "vit_dec_patch_size": 16, | |
| "num_latent_tokens": 32, | |
| "scale_factor": 0.7525, | |
| "finetune_decoder": false, | |
| "is_legacy": false | |
| }, | |
| "maskgen": { | |
| "decoder_embed_dim": 1024, | |
| "decoder_depth": 16, | |
| "decoder_num_heads": 16, | |
| "micro_condition": true, | |
| "micro_condition_embed_dim": 256, | |
| "text_drop_prob": 0.1, | |
| "cfg": 3.0, | |
| "cfg_schedule": "linear", | |
| "num_iter": 32, | |
| "temperature": 1.0, | |
| "sample_aesthetic_score": 6.5 | |
| } | |
| }, | |
| "losses": { | |
| "diffloss_d": 8, | |
| "diffloss_w": 1024 | |
| }, | |
| "dataset": { | |
| "preprocessing": { | |
| "crop_size": 256 | |
| } | |
| } | |
| } |