acceleration:
  load_text_encoder_in_8bit: true
  mixed_precision_mode: bf16
  quantization: int8-quanto
  text_encoder_device: cuda
checkpoints:
  interval: 250
  keep_last_n: -1
  precision: bfloat16
data:
  num_dataloader_workers: 2
  preprocessed_data_root: DELETED
flow_matching:
  timestep_sampling_mode: shifted_logit_normal
  timestep_sampling_params: {}
hub:
  hub_model_id: null
  push_to_hub: false
lora:
  alpha: 128
  dropout: 0.0
  rank: 128
  target_modules:
  - to_k
  - to_q
  - to_v
  - to_out.0
model:
  load_checkpoint: DELETED
  model_path: DELETED
  text_encoder_path: DELETED
  training_mode: lora
optimization:
  batch_size: 1
  enable_gradient_checkpointing: true
  gradient_accumulation_steps: 1
  learning_rate: 0.0001
  max_grad_norm: 1.0
  optimizer_type: adamw8bit
  scheduler_params: {}
  scheduler_type: linear
  steps: 10000
output_dir: DELETED
seed: 42
training_strategy:
  audio_latents_dir: audio_latents
  first_frame_conditioning_p: 0.5
  name: text_to_video
  with_audio: true
validation:
  frame_rate: 25.0
  generate_audio: true
  guidance_scale: 4.0
  images: null
  include_reference_in_output: false
  inference_steps: 50
  interval: 100
  negative_prompt: worst quality, inconsistent motion, blurry, jittery, distorted
  prompts:
  - DELETED
  - DELETED
  reference_downscale_factor: 1
  reference_videos: null
  seed: 42
  skip_initial_validation: true
  stg_blocks:
  - 29
  stg_mode: stg_av
  stg_scale: 0.0
  video_dims:
  - 960
  - 544
  - 89
  videos_per_prompt: 1
wandb:
  enabled: false
  entity: null
  log_validation_videos: true
  project: ltx-2-trainer
  tags:
  - ltx2
  - lora