switch model

app.py

@@ -24,27 +24,18 @@ MODELS = {"RealVisXL 5 Lightning": "SG161222/RealVisXL_V5.0_Lightning",
 def load_model(model_id):
     global pipe, last_loaded_model
 
-    if model_id != last_loaded_model:
-        
+    if model_id != last_loaded_model:        
         # Initialize the models and pipeline
         controlnet = ControlNetUnionModel.from_pretrained(
             "brad-twinkl/controlnet-union-sdxl-1.0-promax", torch_dtype=torch.float16
-        )
-        vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
-        if pipe is not None:
-            optionally_disable_offloading(pipe)
-            torch_gc()
+        ).to(device)
+        vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16).to(device)
         pipe = StableDiffusionXLControlNetTileSRPipeline.from_pretrained(
             MODELS[model_id], controlnet=controlnet, vae=vae, torch_dtype=torch.float16, use_safetensors=True, variant="fp16"
-        )
-        pipe.enable_model_cpu_offload()  # << Enable this if you have limited VRAM
+        ).to(device)
+        #pipe.enable_model_cpu_offload()  # << Enable this if you have limited VRAM
         pipe.enable_vae_tiling() # << Enable this if you have limited VRAM
         pipe.enable_vae_slicing() # << Enable this if you have limited VRAM
-
-        unet = UNet2DConditionModel.from_pretrained(MODELS[model_id], subfolder="unet", variant="fp16", use_safetensors=True)
-        quantize_8bit(unet)  # << Enable this if you have limited VRAM
-        pipe.unet = unet
-
         last_loaded_model = model_id
 
 load_model("RealVisXL 5 Lightning")

pipeline/util.py

@@ -213,7 +213,7 @@ def torch_gc():
     if torch.cuda.is_available():
         with torch.cuda.device("cuda"):
             torch.cuda.empty_cache()
-            #torch.cuda.ipc_collect()
+            torch.cuda.ipc_collect()
 
     gc.collect()