Upload potat1_config.yaml
Browse files- potat1_config.yaml +66 -0
potat1_config.yaml
ADDED
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
pretrained_model_path: /home/ubuntu/models/model_scope_diffusers/
|
2 |
+
output_dir: /home/ubuntu/outputs
|
3 |
+
train_data:
|
4 |
+
width: 1024
|
5 |
+
height: 576
|
6 |
+
use_bucketing: true
|
7 |
+
sample_start_idx: 1
|
8 |
+
fps: 24
|
9 |
+
frame_step: 1
|
10 |
+
n_sample_frames: 20
|
11 |
+
json_path: /home/ubuntu/Video-BLIP2-Preprocessor/train_data/my_videos.json
|
12 |
+
validation_data:
|
13 |
+
prompt: duck
|
14 |
+
sample_preview: false
|
15 |
+
num_frames: 16
|
16 |
+
width: 1024
|
17 |
+
height: 576
|
18 |
+
num_inference_steps: 25
|
19 |
+
guidance_scale: 9
|
20 |
+
dataset_types:
|
21 |
+
- json
|
22 |
+
validation_steps: 500
|
23 |
+
extra_unet_params: null
|
24 |
+
extra_text_encoder_params: null
|
25 |
+
train_batch_size: 1
|
26 |
+
max_train_steps: 50000
|
27 |
+
learning_rate: 5.0e-06
|
28 |
+
scale_lr: false
|
29 |
+
lr_scheduler: constant
|
30 |
+
lr_warmup_steps: 0
|
31 |
+
adam_beta1: 0.9
|
32 |
+
adam_beta2: 0.999
|
33 |
+
adam_weight_decay: 0.01
|
34 |
+
adam_epsilon: 1.0e-08
|
35 |
+
max_grad_norm: 1.0
|
36 |
+
gradient_accumulation_steps: 1
|
37 |
+
checkpointing_steps: 5000
|
38 |
+
resume_from_checkpoint: null
|
39 |
+
mixed_precision: fp16
|
40 |
+
use_8bit_adam: false
|
41 |
+
enable_xformers_memory_efficient_attention: false
|
42 |
+
enable_torch_2_attn: true
|
43 |
+
seed: 64
|
44 |
+
extend_dataset: false
|
45 |
+
cached_latent_dir: null
|
46 |
+
use_unet_lora: false
|
47 |
+
unet_lora_modules:
|
48 |
+
- ResnetBlock2D
|
49 |
+
text_encoder_lora_modules:
|
50 |
+
- CLIPEncoderLayer
|
51 |
+
lora_rank: 16
|
52 |
+
lora_path: ''
|
53 |
+
kwargs: {}
|
54 |
+
cache_latents: false
|
55 |
+
gradient_checkpointing: true
|
56 |
+
offset_noise_strength: 0.1
|
57 |
+
text_encoder_gradient_checkpointing: false
|
58 |
+
train_text_encoder: true
|
59 |
+
trainable_modules:
|
60 |
+
- attn1
|
61 |
+
- attn2
|
62 |
+
- temp_conv
|
63 |
+
trainable_text_modules:
|
64 |
+
- all
|
65 |
+
use_offset_noise: false
|
66 |
+
use_text_lora: false
|