| { | |
| "predictor": { | |
| "_target_": "comit.components.predictor.RuledDiT", | |
| "depth": 12, | |
| "hidden_size": 768, | |
| "in_channels": 4, | |
| "max_input_size": 32, | |
| "mlp_ratio": 4.0, | |
| "msg_tokens_dim": 6, | |
| "num_heads": 12, | |
| "num_msg_tokens": 256, | |
| "out_channels": 4, | |
| "patch_size": 2, | |
| "representation_layer": 4 | |
| }, | |
| "quantizer": { | |
| "_target_": "comit.components.quantizer.FSQ", | |
| "apply_corrupt_tokens_p": 0.0, | |
| "corrupt_tokens_p": 0.0, | |
| "drop_quant_p": 0.0, | |
| "levels": [ | |
| 8, | |
| 8, | |
| 8, | |
| 5, | |
| 5, | |
| 5 | |
| ], | |
| "min_corrupt_tokens_p": null | |
| }, | |
| "vae": { | |
| "_target_": "comit.components.vae.SDAutoencoderKL", | |
| "model_type": "ema" | |
| } | |
| } |