quantize
Browse files- src/pipeline.py +1 -1
src/pipeline.py
CHANGED
|
@@ -50,8 +50,8 @@ def load_pipeline() -> Pipeline:
|
|
| 50 |
)
|
| 51 |
|
| 52 |
pipeline.transformer.to(memory_format=torch.channels_last)
|
| 53 |
-
# quantize_(pipeline.vae, int8_weight_only())
|
| 54 |
pipeline.vae.to(memory_format=torch.channels_last)
|
|
|
|
| 55 |
pipeline.vae = torch.compile(pipeline.vae, mode="max-autotune", fullgraph=True)
|
| 56 |
pipeline.to("cuda")
|
| 57 |
|
|
|
|
| 50 |
)
|
| 51 |
|
| 52 |
pipeline.transformer.to(memory_format=torch.channels_last)
|
|
|
|
| 53 |
pipeline.vae.to(memory_format=torch.channels_last)
|
| 54 |
+
quantize_(pipeline.vae, int8_weight_only())
|
| 55 |
pipeline.vae = torch.compile(pipeline.vae, mode="max-autotune", fullgraph=True)
|
| 56 |
pipeline.to("cuda")
|
| 57 |
|