transformer/config.json
| 1 | { |
| 2 | "_class_name": "LTX2VideoTransformer3DModel", |
| 3 | "_diffusers_version": "0.37.0.dev0", |
| 4 | "activation_fn": "gelu-approximate", |
| 5 | "attention_bias": true, |
| 6 | "attention_head_dim": 128, |
| 7 | "attention_out_bias": true, |
| 8 | "audio_attention_head_dim": 64, |
| 9 | "audio_cross_attention_dim": 2048, |
| 10 | "audio_hop_length": 160, |
| 11 | "audio_in_channels": 128, |
| 12 | "audio_num_attention_heads": 32, |
| 13 | "audio_out_channels": 128, |
| 14 | "audio_patch_size": 1, |
| 15 | "audio_patch_size_t": 1, |
| 16 | "audio_pos_embed_max_pos": 20, |
| 17 | "audio_sampling_rate": 16000, |
| 18 | "audio_scale_factor": 4, |
| 19 | "base_height": 2048, |
| 20 | "base_width": 2048, |
| 21 | "caption_channels": 3840, |
| 22 | "causal_offset": 1, |
| 23 | "cross_attention_dim": 4096, |
| 24 | "cross_attn_timestep_scale_multiplier": 1000, |
| 25 | "in_channels": 128, |
| 26 | "norm_elementwise_affine": false, |
| 27 | "norm_eps": 1e-06, |
| 28 | "num_attention_heads": 32, |
| 29 | "num_layers": 48, |
| 30 | "out_channels": 128, |
| 31 | "patch_size": 1, |
| 32 | "patch_size_t": 1, |
| 33 | "pos_embed_max_pos": 20, |
| 34 | "qk_norm": "rms_norm_across_heads", |
| 35 | "rope_double_precision": true, |
| 36 | "rope_theta": 10000.0, |
| 37 | "rope_type": "split", |
| 38 | "timestep_scale_multiplier": 1000, |
| 39 | "vae_scale_factors": [ |
| 40 | 8, |
| 41 | 32, |
| 42 | 32 |
| 43 | ] |
| 44 | } |
| 45 | |