minimax_h3_video ¶

MiniMaxH3VideoVAEArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), scaling_factor: float = 1.0, temporal_compression_ratio: int = 4, spatial_compression_ratio: int = 16, _class_name: str = 'AutoencoderKLMiniMaxH3', in_channels: int = 3, out_channels: int = 3, latent_channels: int = 24, block_out_channels: tuple[int, ...] = (128, 256, 256, 512, 512, 1024), layers_per_block: int = 2, spatial_downsample_factors: tuple[int, ...] = (2, 2, 2, 2, 1, 1), temporal_downsample_factors: tuple[int, ...] = (1, 2, 2, 1, 1, 1), norm_num_groups: int = 32, norm_eps: float = 1e-06, spatial_padding_mode: str = 'reflect', decoder_num_layers: int = 36, decoder_num_attention_heads: int = 32, decoder_attention_head_dim: int = 64, decoder_num_register_tokens: int = 4, decoder_ffn_mult: int = 4, decoder_rope_theta: float = 100.0, decoder_rope_dim_ratio: float = 0.75, decoder_norm_eps: float = 1e-05, clip_length: int = 17, token_drop: int = 3, latents_mean: tuple[float, ...] = (0.0,) * 24, latents_std: tuple[float, ...] = (1.0,) * 24)

MiniMaxH3VideoVAEConfig(arch_config: MiniMaxH3VideoVAEArchConfig = MiniMaxH3VideoVAEArchConfig(), load_encoder: bool = True, load_decoder: bool = True, tile_sample_min_height: int = 256, tile_sample_min_width: int = 256, tile_sample_min_num_frames: int = 16, tile_sample_stride_height: int = 192, tile_sample_stride_width: int = 192, tile_sample_stride_num_frames: int = 12, blend_num_frames: int = 0, use_tiling: bool = True, use_temporal_tiling: bool = True, use_parallel_tiling: bool = True, use_temporal_scaling_frames: bool = True, tile_sample_min_overlap_height: int = 64, tile_sample_min_overlap_width: int = 64, *, _resolved_attention_backend: AttentionBackendEnum | None = None)