OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

Add basic Gaudi support to benchmarks/dynamo (#145920)

Browse Source 
This PR adds basic Gaudi support to benchmarks/dynamo

Pull Request resolved: https://github.com/pytorch/pytorch/pull/145920
Approved by: https://github.com/eellison
This commit is contained in:
Katarzyna Fojcik 2025-02-26 14:50:22 +00:00 committed by PyTorch MergeBot
parent be830c8b1c
commit edaf9ddeb5
1 changed files with 9 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
benchmarks/dynamo/common.py
View File

@ -2386,6 +2386,8 @@ class BenchmarkRunner:
if current_device == "cuda": if current_device == "cuda":
torch.cuda.reset_peak_memory_stats() torch.cuda.reset_peak_memory_stats()
empty_gpu_cache(current_device) empty_gpu_cache(current_device)
elif current_device == "hpu":
torch.hpu.reset_peak_memory_stats()
t0 = time.perf_counter() t0 = time.perf_counter()
for _ in range(niters): for _ in range(niters):
fn(model, example_inputs) fn(model, example_inputs)
@ -2393,6 +2395,8 @@ class BenchmarkRunner:
latency = t1 - t0 latency = t1 - t0
if current_device == "cuda": if current_device == "cuda":
peak_mem = get_peak_memory() peak_mem = get_peak_memory()
elif current_device == "hpu":
peak_mem = torch.hpu.max_memory_allocated() / 10**9
elif current_device == "cpu": elif current_device == "cpu":
total = psutil.virtual_memory().total total = psutil.virtual_memory().total
percentage = psutil.Process(os.getpid()).memory_percent() percentage = psutil.Process(os.getpid()).memory_percent()
@ -2543,6 +2547,8 @@ class BenchmarkRunner:
if current_device == "cuda": if current_device == "cuda":
torch.cuda.reset_peak_memory_stats() torch.cuda.reset_peak_memory_stats()
empty_gpu_cache(current_device) empty_gpu_cache(current_device)
elif current_device == "hpu":
torch.hpu.reset_peak_memory_stats()
t0 = time.perf_counter() t0 = time.perf_counter()
for _ in range(niters): for _ in range(niters):
fn(model, example_inputs) fn(model, example_inputs)
@ -2550,6 +2556,8 @@ class BenchmarkRunner:
latency = t1 - t0 latency = t1 - t0
if current_device == "cuda": if current_device == "cuda":
peak_mem = get_peak_memory() peak_mem = get_peak_memory()
elif current_device == "hpu":
peak_mem = torch.hpu.max_memory_allocated() / 10**9
elif current_device == "cpu": elif current_device == "cpu":
total = psutil.virtual_memory().total total = psutil.virtual_memory().total
percentage = psutil.Process(os.getpid()).memory_percent() percentage = psutil.Process(os.getpid()).memory_percent()
@ -2870,7 +2878,7 @@ def parse_args(args=None):
help="ID of the benchmark suite partition to be run. Used to divide CI tasks", help="ID of the benchmark suite partition to be run. Used to divide CI tasks",
) )
parser.add_argument( parser.add_argument(
"--devices", "--device", "-d", action="append", help="cpu or cuda" "--devices", "--device", "-d", action="append", help="cpu, cuda or hpu"
) )
parser.add_argument("--device-index", help="CUDA device index") parser.add_argument("--device-index", help="CUDA device index")
parser.add_argument( parser.add_argument(