jokerbit
/

flux-qa-base

Model card Files Files and versions Community

jokerbit commited on Dec 26, 2024

Commit

69e3350

·

verified ·

1 Parent(s): 76d4497

q_ only

Files changed (1) hide show

src/pipeline.py +2 -1

src/pipeline.py CHANGED Viewed

@@ -46,7 +46,8 @@ def load_pipeline() -> Pipeline:
         local_files_only=True,
         torch_dtype=torch.bfloat16,
     ).to("cuda")
-    pipeline.to(memory_format=torch.channels_last)
     # pipeline.transformer.to(memory_format=torch.channels_last)
     # pipeline.vae.to(memory_format=torch.channels_last)

         local_files_only=True,
         torch_dtype=torch.bfloat16,
     ).to("cuda")
+    quantize_(pipeline.vae, int8_weight_only())
+    # pipeline.to(memory_format=torch.channels_last)
     # pipeline.transformer.to(memory_format=torch.channels_last)
     # pipeline.vae.to(memory_format=torch.channels_last)