jokerbit
/

flux-qa-2

Model card Files Files and versions Community

jokerbit commited on Dec 22, 2024

Commit

d79f069

·

verified ·

1 Parent(s): 0d6a48a

1 line less

Files changed (1) hide show

src/pipeline.py +1 -2

src/pipeline.py CHANGED Viewed

@@ -47,9 +47,8 @@ def load_pipeline() -> Pipeline:
         torch_dtype=torch.bfloat16,
     ).to("cuda")
-    pipeline.transformer.to(memory_format=torch.channels_last)
     pipeline.transformer = torch.compile(pipeline.transformer, mode="max-autotune", fullgraph=False)
-    pipeline.vae.to(memory_format=torch.channels_last)
     quantize_(pipeline.vae, int8_weight_only())
     pipeline.vae = torch.compile(pipeline.vae, fullgraph=True, mode="max-autotune")

         torch_dtype=torch.bfloat16,
     ).to("cuda")
+    pipeline.to(memory_format=torch.channels_last)
     pipeline.transformer = torch.compile(pipeline.transformer, mode="max-autotune", fullgraph=False)
     quantize_(pipeline.vae, int8_weight_only())
     pipeline.vae = torch.compile(pipeline.vae, fullgraph=True, mode="max-autotune")