| { | |
| "model_type": "prism", | |
| "transformer": { | |
| "patch_size": [ | |
| 1, | |
| 1 | |
| ], | |
| "attention_head_dim": 128, | |
| "cross_attn_norm": true, | |
| "added_kv_proj_dim": null, | |
| "eps": 1e-06, | |
| "ffn_dim": 8960, | |
| "freq_dim": 256, | |
| "in_channels": 16, | |
| "num_attention_heads": 12, | |
| "num_layers": 30, | |
| "out_channels": 16, | |
| "qk_norm": "rms_norm_across_heads", | |
| "rope_max_seq_len": 1024, | |
| "text_dim": 4096 | |
| }, | |
| "vae_scale_factor_temporal": 4, | |
| "max_text_length": 256 | |
| } |