Upload handler.py

handler.py

@@ -217,13 +217,13 @@ def load_pipeline_fast(repo_id: str, dtype: torch.dtype) -> Any:
     #pipe.transformer.fuse_qkv_projections()
     pipe.transformer.to(memory_format=torch.channels_last)
     #pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune", fullgraph=True)
-    pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune-no-cudagraphs")
+    #pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune-no-cudagraphs")
     if IS_CC90: quantize_(pipe.vae, float8_dynamic_activation_float8_weight(granularity=PerRow()), device="cuda")
     elif IS_CC89: quantize_(pipe.vae, float8_dynamic_activation_float8_weight(), device="cuda")
     #pipe.vae.fuse_qkv_projections()
     pipe.vae.to(memory_format=torch.channels_last)
     #pipe.vae = torch.compile(pipe.vae, mode="max-autotune", fullgraph=True)
-    pipe.vae = torch.compile(pipe.vae, mode="max-autotune-no-cudagraphs")
+    #pipe.vae = torch.compile(pipe.vae, mode="max-autotune-no-cudagraphs")
     return pipe
 
 class EndpointHandler:
@@ -238,11 +238,15 @@ class EndpointHandler:
         elif IS_COMPILE: self.pipeline = load_pipeline_fast(repo_id, dtype)
         elif IS_LVRAM and IS_CC89: self.pipeline = load_pipeline_lowvram(repo_id, dtype)
         else: self.pipeline = load_pipeline_stable(repo_id, dtype)
+        if IS_PARA: apply_cache_on_pipe(self.pipeline, residual_diff_threshold=0.12)
+        self.pipeline.to("cuda")
         if not IS_COMPILE:
             self.pipeline.enable_vae_slicing()
             self.pipeline.enable_vae_tiling()
-            if IS_PARA: apply_cache_on_pipe(self.pipeline, residual_diff_threshold=0.12)
-        self.pipeline.to("cuda")
+        else:
+            print("Compiling pipeline...")
+            self.pipeline.transformer = torch.compile(self.pipeline.transformer, mode="max-autotune-no-cudagraphs")
+            self.pipeline.vae = torch.compile(self.pipeline.vae, mode="max-autotune-no-cudagraphs")
         gc.collect()
         torch.cuda.empty_cache()
         print_vram()