[Platform] Replace torch.cuda.mem_get_info with torch.accelerator.get_memory_info (#44825)

Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Signed-off-by: Kunshang Ji <jikunshang95@gmail.com>
This commit is contained in:
Kunshang Ji
2026-06-30 14:39:52 +08:00
committed by GitHub
parent 0feca7ffa8
commit fb42e5219e
17 changed files with 56 additions and 52 deletions
+4 -4
View File
@@ -172,7 +172,7 @@ class Worker(WorkerBase):
def sleep(self, level: int = 1) -> None:
torch.accelerator.synchronize()
free_bytes_before_sleep = current_platform.mem_get_info()[0]
free_bytes_before_sleep = torch.accelerator.get_memory_info()[0]
# Save the buffers before level 2 sleep
if level == 2:
@@ -187,7 +187,7 @@ class Worker(WorkerBase):
torch.accelerator.synchronize()
deadline = time.monotonic() + (5.0 if current_platform.is_rocm() else 0)
while True:
free_bytes_after_sleep, total = current_platform.mem_get_info()
free_bytes_after_sleep, total = torch.accelerator.get_memory_info()
freed_bytes = free_bytes_after_sleep - free_bytes_before_sleep
if freed_bytes >= 0 or time.monotonic() >= deadline:
break
@@ -459,8 +459,8 @@ class Worker(WorkerBase):
)
# Profile CUDA graph memory if graphs will be captured.
# Skip on ROCm/HIP/XPU as graph pool handles and mem_get_info behave
# differently and can produce incorrect/negative estimates.
# Skip on ROCm/HIP/XPU as graph pool handles and get_memory_info
# behave differently and can produce incorrect/negative estimates.
cudagraph_memory_estimate = 0
if (
current_platform.is_cuda()