fix pre commit

Signed-off-by: princepride <wangzhipeng628@gmail.com>

benchmarks/kernels/mot_linear_benchmarks.py

@@ -64,7 +64,7 @@ def clear_triton_cache():
     """Clear Triton JIT compilation cache and Python/CUDA memory."""
     gc.collect()
     if torch.cuda.is_available():
-        torch.cuda.empty_cache()
+        torch.accelerator.empty_cache()
     try:
         if hasattr(triton, "runtime") and hasattr(triton.runtime, "cache") and hasattr(triton.runtime.cache, "clear"):
             triton.runtime.cache.clear()
@@ -487,18 +487,18 @@ def benchmark_config(
 
     # JIT warmup
     run()
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
 
     # Capture 1 invocations with CUDA Graph
     graph = torch.cuda.CUDAGraph()
     with torch.cuda.graph(graph):
         run()
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
 
     # Warmup replays
     for _ in range(5):
         graph.replay()
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
 
     start_event = torch.cuda.Event(enable_timing=True)
     end_event = torch.cuda.Event(enable_timing=True)
@@ -507,7 +507,7 @@ def benchmark_config(
     for _ in range(num_iters):
         if cache_flusher is not None:
             cache_flusher.zero_()
-        torch.cuda.synchronize()
+        torch.accelerator.synchronize()
 
         start_event.record()
         graph.replay()

tests/diffusion/kernels/mot/test_mot_linear.py

@@ -176,21 +176,21 @@ def _benchmark(fn, warmup: int = 20, iters: int = 100) -> float:
     cache_flusher = torch.empty(int(256 * 1024 * 1024 / 4), dtype=torch.int32, device="cuda")
     for _ in range(warmup):
         fn()
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
 
     t_flush_start = time.perf_counter()
     for _ in range(iters):
         cache_flusher.zero_()
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
     flush_time_total = time.perf_counter() - t_flush_start
 
     #  3. Measure the total time of "flush cache + operator execution"
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
     t_total_start = time.perf_counter()
     for _ in range(iters):
         cache_flusher.zero_()
         fn()
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
     total_time = time.perf_counter() - t_total_start
 
     # 4. Asynchronous subtraction separation

tests/diffusion/kernels/mot/test_mot_norm.py

@@ -51,11 +51,11 @@ def _benchmark(fn, warmup: int = 20, iters: int = 200) -> float:
     """Return mean latency in milliseconds."""
     for _ in range(warmup):
         fn()
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
     t0 = time.perf_counter()
     for _ in range(iters):
         fn()
-    torch.cuda.synchronize()
+    torch.accelerator.synchronize()
     return (time.perf_counter() - t0) / iters * 1000.0