EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Refactor] Cleanup platform (#5566)

Browse Source 
### What this PR does / why we need it?
1. add `COMPILATION_PASS_KEY` constant
2. clean up useless platform interface `empty_cache`, `synchronize`,
`mem_get_info`, `clear_npu_memory`
3. rename `CUSTOM_OP_REGISTERED` to `_CUSTOM_OP_REGISTERED`
4. remove uesless env `VLLM_ENABLE_CUDAGRAPH_GC`

NPUPlatform is the interface called by vLLM. Do not call it inner
vllm-ascend.

### Does this PR introduce _any_ user-facing change?
This PR is just  a cleanup. All CI should pass.

### How was this patch tested?

- vLLM version: v0.13.0
- vLLM main:
7157596103

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2026-01-07 09:25:55 +08:00
committed by GitHub
parent 6ea2afe5fa
commit 1112208052
9 changed files with 79 additions and 217 deletions
Download Patch File Download Diff File Expand all files Collapse all files

109
tests/ut/test_platform.py
View File

@@ -120,115 +120,6 @@ class TestNPUPlatform(TestBase):
self.assertIsNone(self.platform.inference_mode())
mock_inference_mode.assert_called_once()
@patch("torch.npu.set_device")
def test_set_device_normal(self, mock_set_device):
device = torch.device("npu:0")
self.platform.set_device(device)
mock_set_device.assert_called_once_with(device)
@patch("torch.npu.set_device",
side_effect=RuntimeError("Device not available"))
def test_set_device_failure(self, mock_set_device):
device = torch.device("npu:0")
with self.assertRaises(RuntimeError):
self.platform.set_device(device)
mock_set_device.assert_called_once_with(device)
@patch("torch.npu.empty_cache")
def test_empty_cache_normal(self, mock_empty_cache):
self.platform.empty_cache()
mock_empty_cache.assert_called_once()
@patch("torch.npu.empty_cache",
side_effect=RuntimeError("Cache clearing failed"))
def test_empty_cache_failure(self, mock_empty_cache):
with self.assertRaises(RuntimeError):
self.platform.empty_cache()
mock_empty_cache.assert_called_once()
@patch("torch.npu.synchronize")
def test_synchronize_normal(self, mock_synchronize):
self.platform.synchronize()
mock_synchronize.assert_called_once()
@patch("torch.npu.synchronize",
side_effect=RuntimeError("Synchronization failed"))
def test_synchronize_failure(self, mock_synchronize):
with self.assertRaises(RuntimeError):
self.platform.synchronize()
mock_synchronize.assert_called_once()
@patch("torch.npu.mem_get_info")
def test_mem_get_info_normal(self, mock_mem_get_info):
free_memory_size = 1024
total_memory_size = 2048
memory_info = (free_memory_size, total_memory_size)
mock_mem_get_info.return_value = memory_info
result = self.platform.mem_get_info()
self.assertIsInstance(result, tuple)
self.assertEqual(len(result), 2)
self.assertEqual(result, memory_info)
mock_mem_get_info.assert_called_once()
@patch("torch.npu.mem_get_info",
side_effect=RuntimeError("NPU not available"))
def test_mem_get_info_failure(self, mock_mem_get_info):
with self.assertRaises(RuntimeError):
self.platform.mem_get_info()
mock_mem_get_info.assert_called_once()
@patch("gc.collect")
@patch("torch.npu.empty_cache")
@patch("torch.npu.reset_peak_memory_stats")
def test_clear_npu_memory_normal(self, mock_reset_stats, mock_empty_cache,
mock_gc_collect):
self.platform.clear_npu_memory()
mock_gc_collect.assert_called_once()
mock_empty_cache.assert_called_once()
mock_reset_stats.assert_called_once()
@patch("gc.collect", side_effect=Exception("GC failed"))
@patch("torch.npu.empty_cache")
@patch("torch.npu.reset_peak_memory_stats")
def test_clear_npu_memory_gc_collect_failure(self, mock_reset_stats,
mock_empty_cache,
mock_gc_collect):
with self.assertRaises(Exception):
self.platform.clear_npu_memory()
mock_gc_collect.assert_called_once()
mock_empty_cache.assert_not_called()
mock_reset_stats.assert_not_called()
@patch("gc.collect")
@patch("torch.npu.empty_cache",
side_effect=RuntimeError("Cache clear failed"))
@patch("torch.npu.reset_peak_memory_stats")
def test_clear_npu_memory_empty_cache_failure(self, mock_reset_stats,
mock_empty_cache,
mock_gc_collect):
with self.assertRaises(RuntimeError):
self.platform.clear_npu_memory()
mock_gc_collect.assert_called_once()
mock_empty_cache.assert_called_once()
mock_reset_stats.assert_not_called()
@patch("gc.collect")
@patch("torch.npu.empty_cache")
@patch("torch.npu.reset_peak_memory_stats",
side_effect=RuntimeError("Reset failed"))
def test_clear_npu_memory_reset_stats_failure(self, mock_reset_stats,
mock_empty_cache,
mock_gc_collect):
with self.assertRaises(RuntimeError):
self.platform.clear_npu_memory()
mock_gc_collect.assert_called_once()
mock_empty_cache.assert_called_once()
mock_reset_stats.assert_called_once()
@patch("vllm_ascend.ascend_config.init_ascend_config")
@patch("vllm_ascend.utils.update_aclgraph_sizes")
@patch('vllm_ascend.utils.get_ascend_device_type',