vllm_omni.diffusion.models.ltx2.pipeline_ltx2 ¶

attention_kwargs

audio_hop_length = (
    self.audio_vae.config.mel_hop_length
    if getattr(self, "audio_vae", None) is not None
    else 160
)

audio_sampling_rate = (
    self.audio_vae.config.sample_rate
    if getattr(self, "audio_vae", None) is not None
    else 16000
)

audio_vae = from_pretrained_with_prefetch(
    AutoencoderKLLTX2Audio.from_pretrained,
    model,
    subfolder="audio_vae",
    prefetch_list=ltx2_subfolders,
    local_files_only=local_files_only,
    torch_dtype=dtype,
).to(self.device)

audio_vae_mel_compression_ratio = (
    self.audio_vae.mel_compression_ratio
    if getattr(self, "audio_vae", None) is not None
    else 4
)

audio_vae_temporal_compression_ratio = (
    self.audio_vae.temporal_compression_ratio
    if getattr(self, "audio_vae", None) is not None
    else 4
)

connectors = from_pretrained_with_prefetch(
    LTX2TextConnectors.from_pretrained,
    model,
    subfolder="connectors",
    prefetch_list=ltx2_subfolders,
    local_files_only=local_files_only,
    torch_dtype=dtype,
).to(self.device)

current_timestep

device = get_local_device()

do_classifier_free_guidance

dummy_run_num_frames = 2

guidance_rescale

guidance_scale

interrupt

num_timesteps

od_config = od_config

scheduler = FlowMatchEulerDiscreteScheduler.from_pretrained(
    model,
    subfolder="scheduler",
    local_files_only=local_files_only,
)

text_encoder = from_pretrained_with_prefetch(
    Gemma3ForConditionalGeneration.from_pretrained,
    model,
    subfolder="text_encoder",
    prefetch_list=ltx2_subfolders,
    local_files_only=local_files_only,
    torch_dtype=dtype,
).to(self.device)

tokenizer = AutoTokenizer.from_pretrained(
    model,
    subfolder="tokenizer",
    local_files_only=local_files_only,
)

tokenizer_max_length = int(tokenizer_max_length)

transformer = create_transformer_from_config(
    transformer_config, quant_config=quant_config
)

transformer_spatial_patch_size = (
    self.transformer.config.patch_size
    if getattr(self, "transformer", None) is not None
    else 1
)

transformer_temporal_patch_size = (
    self.transformer.config.patch_size_t
    if getattr(self, "transformer", None) is not None
    else 1
)

vae = from_pretrained_with_prefetch(
    AutoencoderKLLTX2Video.from_pretrained,
    model,
    subfolder="vae",
    prefetch_list=ltx2_subfolders,
    local_files_only=local_files_only,
    torch_dtype=dtype,
).to(self.device)

vae_spatial_compression_ratio = (
    self.vae.spatial_compression_ratio
    if getattr(self, "vae", None) is not None
    else 32
)

vae_temporal_compression_ratio = (
    self.vae.temporal_compression_ratio
    if getattr(self, "vae", None) is not None
    else 8
)

video_processor = VideoProcessor(
    vae_scale_factor=self.vae_spatial_compression_ratio
)

vocoder = from_pretrained_with_prefetch(
    LTX2Vocoder.from_pretrained,
    model,
    subfolder="vocoder",
    prefetch_list=ltx2_subfolders,
    local_files_only=local_files_only,
    torch_dtype=dtype,
).to(self.device)

weights_sources = [
    DiffusersPipelineLoader.ComponentSource(
        model_or_path=od_config.model,
        subfolder="transformer",
        revision=None,
        prefix="transformer.",
        fall_back_to_pt=True,
    )
]

device = get_local_device()

distilled = False

dtype = getattr(od_config, 'dtype', torch.bfloat16)

dummy_run_num_frames = 2

lora_manager = DiffusionLoRAManager(
    pipeline=self.pipe,
    device=self.device,
    dtype=self.dtype,
    max_cached_adapters=od_config.max_cpu_loras,
)

model_path = od_config.model

pipe = LTX2Pipeline(od_config=od_config, prefix=prefix)

upsample_pipe = LTX2LatentUpsamplePipeline(
    vae=self.pipe.vae, od_config=od_config
)

weights_sources = [
    DiffusersPipelineLoader.ComponentSource(
        model_or_path=od_config.model,
        subfolder="transformer",
        revision=None,
        prefix="pipe.transformer.",
        fall_back_to_pt=True,
    )
]