Upload handler.py

handler.py

@@ -67,8 +67,8 @@ def load_pipeline_lowvram(repo_id: str, dtype: torch.dtype) -> Any:
     vae = AutoencoderKL.from_pretrained(repo_id, subfolder="vae", torch_dtype=dtype, quantization_config=float8_config)
     transformer = FluxTransformer2DModel.from_pretrained(repo_id, subfolder="transformer", torch_dtype=dtype, quantization_config=float8_config)
     pipe = FluxPipeline.from_pretrained(repo_id, vae=vae, transformer=transformer, text_encoder_2=load_te2(repo_id, dtype), torch_dtype=dtype, quantization_config=int4_config)
-    pipe.transformer.fuse_qkv_projections()
-    pipe.vae.fuse_qkv_projections()
+    #pipe.transformer.fuse_qkv_projections()
+    #pipe.vae.fuse_qkv_projections()
     pipe.to("cuda")
     return pipe