pretrained_model_path: /home/ubuntu/models/model_scope_diffusers/
output_dir: /home/ubuntu/outputs
train_data:
  width: 1024
  height: 576
  use_bucketing: true
  sample_start_idx: 1
  fps: 24
  frame_step: 1
  n_sample_frames: 20
  json_path: /home/ubuntu/Video-BLIP2-Preprocessor/train_data/my_videos.json
validation_data:
  prompt: duck
  sample_preview: false
  num_frames: 16
  width: 1024
  height: 576
  num_inference_steps: 25
  guidance_scale: 9
dataset_types:
- json
validation_steps: 500
extra_unet_params: null
extra_text_encoder_params: null
train_batch_size: 1
max_train_steps: 50000
learning_rate: 5.0e-06
scale_lr: false
lr_scheduler: constant
lr_warmup_steps: 0
adam_beta1: 0.9
adam_beta2: 0.999
adam_weight_decay: 0.01
adam_epsilon: 1.0e-08
max_grad_norm: 1.0
gradient_accumulation_steps: 1
checkpointing_steps: 5000
resume_from_checkpoint: null
mixed_precision: fp16
use_8bit_adam: false
enable_xformers_memory_efficient_attention: false
enable_torch_2_attn: true
seed: 64
extend_dataset: false
cached_latent_dir: null
use_unet_lora: false
unet_lora_modules:
- ResnetBlock2D
text_encoder_lora_modules:
- CLIPEncoderLayer
lora_rank: 16
lora_path: ''
kwargs: {}
cache_latents: false
gradient_checkpointing: true
offset_noise_strength: 0.1
text_encoder_gradient_checkpointing: false
train_text_encoder: true
trainable_modules:
- attn1
- attn2
- temp_conv
trainable_text_modules:
- all
use_offset_noise: false
use_text_lora: false