[Core][AMD] Migrate fully transparent sleep mode to ROCm platform (#12695)

Signed-off-by: Hollow Man <hollowman@opensuse.org>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: kliuae <kuanfu.liu@embeddedllm.com>
This commit is contained in:
ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟
2025-11-13 01:24:12 +02:00
committed by GitHub
parent 10f01d5a3a
commit 4ca5cd5740
11 changed files with 582 additions and 31 deletions

View File

@@ -8,12 +8,13 @@ import torch
from vllm import LLM, AsyncEngineArgs, AsyncLLMEngine, SamplingParams
from vllm.device_allocator.cumem import CuMemAllocator
from vllm.platforms import current_platform
from vllm.utils.mem_constants import GiB_bytes
from ..utils import create_new_process_for_each_test
@create_new_process_for_each_test()
@create_new_process_for_each_test("fork" if not current_platform.is_rocm() else "spawn")
def test_python_error():
"""
Test if Python error occurs when there's low-level
@@ -39,7 +40,7 @@ def test_python_error():
allocator.wake_up()
@create_new_process_for_each_test()
@create_new_process_for_each_test("fork" if not current_platform.is_rocm() else "spawn")
def test_basic_cumem():
# some tensors from default memory pool
shape = (1024, 1024)
@@ -72,7 +73,7 @@ def test_basic_cumem():
assert torch.allclose(output, torch.ones_like(output) * 3)
@create_new_process_for_each_test()
@create_new_process_for_each_test("fork" if not current_platform.is_rocm() else "spawn")
def test_cumem_with_cudagraph():
allocator = CuMemAllocator.get_instance()
with allocator.use_memory_pool():
@@ -117,7 +118,7 @@ def test_cumem_with_cudagraph():
assert torch.allclose(y, x + 1)
@create_new_process_for_each_test()
@create_new_process_for_each_test("fork" if not current_platform.is_rocm() else "spawn")
@pytest.mark.parametrize(
"model",
[