[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
+6 -6
View File
@@ -6527,7 +6527,7 @@ class GPUModelRunner(
mem_samples: list[int] = []
for i, desc in enumerate(profile_descs):
mem_before = torch.cuda.mem_get_info()[0]
mem_before = torch.accelerator.get_memory_info()[0]
self._warmup_and_capture(
desc,
cudagraph_runtime_mode=mode,
@@ -6541,7 +6541,7 @@ class GPUModelRunner(
),
)
torch.accelerator.synchronize()
free_after = torch.cuda.mem_get_info()[0]
free_after = torch.accelerator.get_memory_info()[0]
mem_samples.append(mem_before - free_after)
first_capture = mem_samples[0]
@@ -6563,10 +6563,10 @@ class GPUModelRunner(
)
if encoder_cudagraph_manager is not None:
mem_before = torch.cuda.mem_get_info()[0]
mem_before = torch.accelerator.get_memory_info()[0]
encoder_cudagraph_manager.capture(graph_pool=encoder_profiling_pool)
torch.accelerator.synchronize()
free_after = torch.cuda.mem_get_info()[0]
free_after = torch.accelerator.get_memory_info()[0]
encoder_memory_estimate = max(mem_before - free_after, 0)
logger.debug(
@@ -6632,7 +6632,7 @@ class GPUModelRunner(
with self._freeze_gc(), graph_capture(device=self.device):
torch.accelerator.synchronize()
torch.accelerator.empty_cache()
start_free_gpu_memory = torch.cuda.mem_get_info()[0]
start_free_gpu_memory = torch.accelerator.get_memory_info()[0]
for (
runtime_mode,
@@ -6650,7 +6650,7 @@ class GPUModelRunner(
self.encoder_cudagraph_manager.capture(graph_pool=encoder_graph_pool)
torch.accelerator.synchronize()
end_free_gpu_memory = torch.cuda.mem_get_info()[0]
end_free_gpu_memory = torch.accelerator.get_memory_info()[0]
# Disable cudagraph capturing globally, so any unexpected cudagraph
# capturing will be detected and raise an error after here.