Skip to content

Updating instructions to load 2-layer Wan model #712

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
tv-karthikeya wants to merge 1 commit into from
Closed

Updating instructions to load 2-layer Wan model #712

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
35 changes: 12 additions & 23 deletions examples/diffusers/wan/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,8 +109,8 @@ python wan_lightning.py

```python
# Reduce to 2 layers for faster inference
pipeline.transformer.model.transformer_high.config.num_layers = 2
pipeline.transformer.model.transformer_low.config.num_layers = 2
pipeline.transformer.model.transformer_high.config['num_layers'] = 2
pipeline.transformer.model.transformer_low.config['num_layers']= 2

original_blocks = pipeline.transformer.model.transformer_high.blocks
org_blocks = pipeline.transformer.model.transformer_low.blocks
Expand Down Expand Up @@ -161,26 +161,18 @@ The configuration includes dual specializations for WAN's high and low noise mod
"transformer": {
"specializations":[
{
"batch_size":"1",
"cl":"5040",
"latent_height":"24",
"latent_width":"40",
"model_type":"1",
"num_channels":"16",
"num_frames":"21",
"sequence_length":"512",
"steps":"1"
"batch_size": "1",
"num_channels": "16",
"steps": "1",
"sequence_length": "512",
"model_type": "1"
},
{
"batch_size":"1",
"cl":"5040",
"latent_height":"24",
"latent_width":"40",
"model_type":"2",
"num_channels":"16",
"num_frames":"21",
"sequence_length":"512",
"steps":"1"
"batch_size": "1",
"num_channels": "16",
"steps": "1",
"sequence_length": "512",
"model_type": "2"
}
]
}
Expand All @@ -192,9 +184,6 @@ The configuration includes dual specializations for WAN's high and low noise mod
#### Specializations
- `batch_size`: Batch size for inference
- `num_channels`: Number of latent channels (16 for WAN)
- `num_frames`: Number of latent frames (21 for 81 input frames)
- `latent_height`/`latent_width`: Latent space dimensions
- `cl`: Compressed latent dimension for transformer
- `sequence_length` : Sequence length of text encoder 512
- `model_type`: 1 for high noise model, 2 for low noise model

Expand Down
4 changes: 2 additions & 2 deletions examples/diffusers/wan/wan_lightning_custom.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,8 +85,8 @@ def load_wan_lora(path: str):
# Uncomment the following lines to use only a subset of transformer layers:
#
# # Configure for 2-layer model (faster inference)
# pipeline.transformer.model.transformer_high.config.num_layers = 1
# pipeline.transformer.model.transformer_low.config.num_layers = 1
# pipeline.transformer.model.transformer_high.config['num_layers'] = 2
# pipeline.transformer.model.transformer_low.config['num_layers']= 2
#
# # Reduce high noise transformer blocks
# original_blocks = pipeline.transformer.model.transformer_high.blocks
Expand Down