Update app.py

app.py

@@ -31,7 +31,7 @@ torch.set_float32_matmul_precision("high")
 torch._inductor.config.conv_1x1_as_mm = True
 torch._inductor.config.coordinate_descent_tuning = True
 torch._inductor.config.epilogue_fusion = False
-torch._inductor.config.coordinate_descent_check_all_directions = True
+torch._inductor.config.coordinate_descent_check_all_directions = False
 
 # Load LoRAs from JSON file
 with open('loras.json', 'r') as f:
@@ -52,7 +52,7 @@ torch_dtype=torch.bfloat16,
 trust_remote_code=True,
 ).to(device)
 
-clipmodel = 'norm'
+clipmodel = 'long'
 if clipmodel == "long":
     model_id = "zer0int/LongCLIP-GmP-ViT-L-14"
     config = CLIPConfig.from_pretrained(model_id)
@@ -68,12 +68,14 @@ pipe.tokenizer = clip_processor.tokenizer
 pipe.text_encoder = clip_model.text_model
 pipe.tokenizer_max_length = maxtokens
 pipe.text_encoder.dtype = torch.bfloat16
+pipe.vae = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype).to("cuda")
+
 
 pipe.transformer.to(memory_format=torch.channels_last)
-pipe.vae.to(memory_format=torch.channels_last)
+#pipe.vae.to(memory_format=torch.channels_last)
 
-pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune", fullgraph=True)
-pipe.vae.decode = torch.compile(pipe.vae.decode, mode="max-autotune", fullgraph=True)
+pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune", fullgraph=False)
+#pipe.vae.decode = torch.compile(pipe.vae.decode, mode="max-autotune", fullgraph=False)
 
 MAX_SEED = 2**32-1