[misc] Remove unit test cases that fail the admission criteria (#30690)
This commit is contained in:
Liangsheng Yin
@@ -10,8 +10,8 @@ from unittest.mock import MagicMock, patch
|
||||
import torch
|
||||
|
||||
from sglang.srt.platforms import _load_platform_class, _resolve_platform
|
||||
from sglang.srt.platforms.cpu import CpuDeviceMixin, CpuSRTPlatform
|
||||
from sglang.srt.platforms.cuda import CudaDeviceMixin, CudaSRTPlatform
|
||||
from sglang.srt.platforms.cpu import CpuSRTPlatform
|
||||
from sglang.srt.platforms.cuda import CudaSRTPlatform
|
||||
from sglang.srt.platforms.device_mixin import (
|
||||
CpuArchEnum,
|
||||
DeviceCapability,
|
||||
@@ -47,41 +47,6 @@ def _make_device_mixin(enum, name, dtype):
|
||||
return M()
|
||||
|
||||
|
||||
class _StubPlatform(SRTPlatform):
|
||||
"""Concrete SRTPlatform with minimal defaults for testing overrides."""
|
||||
|
||||
_enum = PlatformEnum.CUDA
|
||||
device_name = "cuda"
|
||||
device_type = "cuda"
|
||||
|
||||
def get_device_total_memory(self, device_id=0):
|
||||
return 10**9
|
||||
|
||||
def get_current_memory_usage(self, device=None):
|
||||
return 5 * 10**8
|
||||
|
||||
def get_default_attention_backend(self):
|
||||
return "flashinfer"
|
||||
|
||||
def get_graph_runner_cls(self):
|
||||
return object
|
||||
|
||||
def get_mha_kv_pool_cls(self):
|
||||
return object
|
||||
|
||||
def get_mla_kv_pool_cls(self):
|
||||
return object
|
||||
|
||||
def get_dsa_kv_pool_cls(self):
|
||||
return object
|
||||
|
||||
def get_paged_allocator_cls(self):
|
||||
return object
|
||||
|
||||
def get_piecewise_backend_cls(self):
|
||||
return object
|
||||
|
||||
|
||||
def _make_platform_ep(name, load_fn=None):
|
||||
"""Create a mock entry point for platform plugins."""
|
||||
ep = MagicMock()
|
||||
@@ -240,78 +205,12 @@ class TestCudaDeviceMixin(CustomTestCase):
|
||||
base = CudaSRTPlatform()
|
||||
self.assertEqual(base.get_device(2), torch.device("cuda", 2))
|
||||
|
||||
def test_cuda_platform_identity(self):
|
||||
base = CudaSRTPlatform()
|
||||
self.assertTrue(base.is_cuda())
|
||||
self.assertTrue(base.is_cuda_alike())
|
||||
self.assertIsInstance(base, CudaDeviceMixin)
|
||||
|
||||
@patch("torch.cuda.get_device_properties")
|
||||
def test_default_get_device_total_memory_uses_cuda(
|
||||
self, mock_get_device_properties
|
||||
):
|
||||
mock_get_device_properties.return_value.total_memory = 123
|
||||
base = CudaSRTPlatform()
|
||||
self.assertEqual(base.get_device_total_memory(1), 123)
|
||||
mock_get_device_properties.assert_called_once_with(1)
|
||||
|
||||
@patch("torch.cuda.max_memory_allocated", return_value=456)
|
||||
def test_default_get_current_memory_usage_uses_cuda(
|
||||
self, mock_max_memory_allocated
|
||||
):
|
||||
base = CudaSRTPlatform()
|
||||
device = torch.device("cuda", 1)
|
||||
self.assertEqual(base.get_current_memory_usage(device), 456.0)
|
||||
mock_max_memory_allocated.assert_called_once_with(device)
|
||||
|
||||
@patch("torch.cuda.set_device")
|
||||
def test_default_set_device_uses_cuda(self, mock_set_device):
|
||||
base = CudaSRTPlatform()
|
||||
device = torch.device("cuda", 1)
|
||||
base.set_device(device)
|
||||
mock_set_device.assert_called_once_with(device)
|
||||
|
||||
@patch("torch.cuda.get_device_name", return_value="NVIDIA H100")
|
||||
def test_default_get_device_name_uses_cuda(self, mock_get_device_name):
|
||||
base = CudaSRTPlatform()
|
||||
self.assertEqual(base.get_device_name(1), "NVIDIA H100")
|
||||
mock_get_device_name.assert_called_once_with(1)
|
||||
|
||||
@patch("torch.cuda.get_device_properties")
|
||||
def test_default_get_device_uuid_uses_cuda(self, mock_get_device_properties):
|
||||
mock_get_device_properties.return_value.uuid = "1234"
|
||||
base = CudaSRTPlatform()
|
||||
self.assertEqual(base.get_device_uuid(1), "1234")
|
||||
mock_get_device_properties.assert_called_once_with(1)
|
||||
|
||||
@patch("torch.cuda.get_device_capability", return_value=(9, 0))
|
||||
def test_default_get_device_capability_uses_cuda(self, mock_get_device_capability):
|
||||
base = CudaSRTPlatform()
|
||||
self.assertEqual(base.get_device_capability(1), DeviceCapability(9, 0))
|
||||
mock_get_device_capability.assert_called_once_with(1)
|
||||
|
||||
@patch("torch.cuda.empty_cache")
|
||||
def test_default_empty_cache_uses_cuda(self, mock_empty_cache):
|
||||
base = CudaSRTPlatform()
|
||||
base.empty_cache()
|
||||
mock_empty_cache.assert_called_once_with()
|
||||
|
||||
@patch("torch.cuda.synchronize")
|
||||
def test_default_synchronize_uses_cuda(self, mock_synchronize):
|
||||
base = CudaSRTPlatform()
|
||||
base.synchronize()
|
||||
mock_synchronize.assert_called_once_with()
|
||||
|
||||
@patch("torch.cuda.mem_get_info", return_value=(123, 456), create=True)
|
||||
def test_default_get_available_memory_uses_cuda(self, mock_mem_get_info):
|
||||
base = CudaSRTPlatform()
|
||||
self.assertEqual(base.get_available_memory(1), (123, 456))
|
||||
mock_mem_get_info.assert_called_once_with(1)
|
||||
|
||||
def test_default_distributed_backend_is_nccl(self):
|
||||
base = CudaSRTPlatform()
|
||||
self.assertEqual(base.get_torch_distributed_backend_str(), "nccl")
|
||||
|
||||
@patch("torch.cuda.manual_seed_all")
|
||||
@patch("torch.manual_seed")
|
||||
@patch("sglang.srt.platforms.device_mixin.np.random.seed")
|
||||
@@ -335,32 +234,12 @@ class TestCudaDeviceMixin(CustomTestCase):
|
||||
class TestCpuDeviceMixin(CustomTestCase):
|
||||
"""Tests for CPU device operation defaults (covers both x86 and ARM)."""
|
||||
|
||||
def test_cpu_platform_identity(self):
|
||||
base = CpuSRTPlatform()
|
||||
self.assertTrue(base.is_cpu())
|
||||
self.assertFalse(base.is_cuda())
|
||||
self.assertFalse(base.is_cuda_alike())
|
||||
self.assertIsInstance(base, CpuDeviceMixin)
|
||||
|
||||
def test_default_get_device_returns_cpu_device(self):
|
||||
base = CpuSRTPlatform()
|
||||
# ``local_rank`` is ignored — CPU has no per-rank device.
|
||||
self.assertEqual(base.get_device(0), torch.device("cpu"))
|
||||
self.assertEqual(base.get_device(7), torch.device("cpu"))
|
||||
|
||||
@patch("sglang.srt.platforms.cpu.psutil.virtual_memory")
|
||||
def test_default_get_device_total_memory_uses_psutil(self, mock_vm):
|
||||
mock_vm.return_value.total = 12345
|
||||
base = CpuSRTPlatform()
|
||||
self.assertEqual(base.get_device_total_memory(), 12345)
|
||||
|
||||
@patch("sglang.srt.platforms.cpu.psutil.virtual_memory")
|
||||
def test_default_get_available_memory_uses_psutil(self, mock_vm):
|
||||
mock_vm.return_value.available = 100
|
||||
mock_vm.return_value.total = 200
|
||||
base = CpuSRTPlatform()
|
||||
self.assertEqual(base.get_available_memory(), (100, 200))
|
||||
|
||||
@patch("sglang.srt.platforms.cpu.psutil.virtual_memory")
|
||||
def test_default_get_current_memory_usage_is_system_used(self, mock_vm):
|
||||
mock_vm.return_value.total = 1000
|
||||
@@ -378,14 +257,6 @@ class TestCpuDeviceMixin(CustomTestCase):
|
||||
free = base.get_device_total_memory() - base.get_current_memory_usage()
|
||||
self.assertEqual(free, 300)
|
||||
|
||||
@patch("torch.cpu.set_device")
|
||||
def test_default_set_device_uses_torch_cpu(self, mock_set_device):
|
||||
base = CpuSRTPlatform()
|
||||
device = torch.device("cpu")
|
||||
base.set_device(device)
|
||||
# Documented CPU no-op, but called for symmetry with CudaDeviceMixin.
|
||||
mock_set_device.assert_called_once_with(device)
|
||||
|
||||
def test_default_set_device_does_not_flip_default(self):
|
||||
base = CpuSRTPlatform()
|
||||
# Must not call torch.set_default_device — process-wide default stays put.
|
||||
@@ -394,22 +265,6 @@ class TestCpuDeviceMixin(CustomTestCase):
|
||||
after = torch.empty(0).device
|
||||
self.assertEqual(before, after)
|
||||
|
||||
@patch("sglang.srt.platforms.cpu.gc.collect")
|
||||
def test_default_empty_cache_calls_gc_collect(self, mock_collect):
|
||||
base = CpuSRTPlatform()
|
||||
base.empty_cache()
|
||||
mock_collect.assert_called_once_with()
|
||||
|
||||
@patch("torch.cpu.synchronize")
|
||||
def test_default_synchronize_uses_torch_cpu(self, mock_synchronize):
|
||||
base = CpuSRTPlatform()
|
||||
base.synchronize()
|
||||
mock_synchronize.assert_called_once_with()
|
||||
|
||||
def test_default_distributed_backend_is_gloo(self):
|
||||
base = CpuSRTPlatform()
|
||||
self.assertEqual(base.get_torch_distributed_backend_str(), "gloo")
|
||||
|
||||
@patch("platform.machine", return_value="aarch64")
|
||||
def test_cpu_arch_property_resolves_and_caches(self, mock_machine):
|
||||
base = CpuSRTPlatform()
|
||||
@@ -431,15 +286,6 @@ class TestCpuDeviceMixin(CustomTestCase):
|
||||
name = base.get_device_name()
|
||||
self.assertIn("x86_64", name)
|
||||
|
||||
@patch("platform.machine", return_value="aarch64")
|
||||
def test_get_device_uuid_returns_machine(self, _mock_machine):
|
||||
base = CpuSRTPlatform()
|
||||
self.assertEqual(base.get_device_uuid(), "aarch64")
|
||||
|
||||
def test_get_device_capability_returns_none(self):
|
||||
base = CpuSRTPlatform()
|
||||
self.assertIsNone(base.get_device_capability())
|
||||
|
||||
def test_cpu_srt_platform_capabilities(self):
|
||||
base = CpuSRTPlatform()
|
||||
self.assertFalse(base.supports_fp8())
|
||||
@@ -449,32 +295,6 @@ class TestCpuDeviceMixin(CustomTestCase):
|
||||
self.assertFalse(base.is_pin_memory_available())
|
||||
|
||||
|
||||
class TestSRTPlatformOverrides(CustomTestCase):
|
||||
"""Tests for SRTPlatform method overrides via plugins."""
|
||||
|
||||
def test_custom_get_dispatch_key_name(self):
|
||||
class P(_StubPlatform):
|
||||
_enum = PlatformEnum.NPU
|
||||
device_name = "npu"
|
||||
device_type = "npu"
|
||||
|
||||
def get_dispatch_key_name(self):
|
||||
return "npu"
|
||||
|
||||
self.assertEqual(P().get_dispatch_key_name(), "npu")
|
||||
|
||||
def test_custom_get_compile_backend(self):
|
||||
class P(_StubPlatform):
|
||||
_enum = PlatformEnum.NPU
|
||||
device_name = "npu"
|
||||
device_type = "npu"
|
||||
|
||||
def get_compile_backend(self, mode=None):
|
||||
return "inductor"
|
||||
|
||||
self.assertEqual(P().get_compile_backend(mode="npugraph_ex"), "inductor")
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Platform Discovery: _resolve_platform
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
Reference in New Issue
Block a user