ltx2

LTX-2 Transformer configuration for native FastVideo integration.

Classes

fastvideo.configs.models.dits.ltx2.LTX2VideoArchConfig dataclass

LTX2VideoArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), _fsdp_shard_conditions: list = (lambda: [is_ltx2_blocks])(), _compile_conditions: list = (lambda: [is_ltx2_blocks])(), param_names_mapping: dict = (lambda: {'^model\\.diffusion_model\\.(.*)$': 'model.\\1', '^diffusion_model\\.(.*)$': 'model.\\1', '^model\\.(.*)$': 'model.\\1', '^(.*)$': 'model.\\1'})(), reverse_param_names_mapping: dict = (lambda: {})(), lora_param_names_mapping: dict = (lambda: {})(), _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (SLIDING_TILE_ATTN, SAGE_ATTN, FLASH_ATTN, TORCH_SDPA, VIDEO_SPARSE_ATTN, VMOBA_ATTN, SAGE_ATTN_THREE, SLA_ATTN, SAGE_SLA_ATTN), hidden_size: int = 0, num_attention_heads: int = 32, num_channels_latents: int = 128, in_channels: int | None = None, out_channels: int | None = None, exclude_lora_layers: list[str] = list(), boundary_ratio: float | None = None, attention_head_dim: int = 128, num_layers: int = 48, cross_attention_dim: int = 4096, caption_channels: int = 3840, norm_eps: float = 1e-06, attention_type: str = 'default', rope_type: str = 'split', double_precision_rope: bool = True, positional_embedding_theta: float = 10000.0, positional_embedding_max_pos: list[int] = (lambda: [20, 2048, 2048])(), timestep_scale_multiplier: int = 1000, use_middle_indices_grid: bool = True, patch_size: tuple[int, int, int] = (1, 1, 1), audio_num_attention_heads: int = 32, audio_attention_head_dim: int = 64, audio_in_channels: int = 128, audio_out_channels: int = 128, audio_cross_attention_dim: int = 2048, audio_positional_embedding_max_pos: list[int] = (lambda: [20])(), av_ca_timestep_scale_multiplier: int = 1)

Bases: DiTArchConfig

Architecture configuration for LTX-2 video transformer.

fastvideo.configs.models.dits.ltx2.LTX2VideoConfig dataclass

LTX2VideoConfig(arch_config: DiTArchConfig = LTX2VideoArchConfig(), prefix: str = 'ltx2', quant_config: QuantizationConfig | None = None)

Bases: DiTConfig

Main configuration for LTX-2 transformer.

Functions

fastvideo.configs.models.dits.ltx2.is_ltx2_blocks

is_ltx2_blocks(name: str, _module) -> bool

FSDP shard condition for LTX-2 transformer blocks.

Source code in fastvideo/configs/models/dits/ltx2.py

def is_ltx2_blocks(name: str, _module) -> bool:
    """FSDP shard condition for LTX-2 transformer blocks."""
    return "transformer_blocks" in name