[AMD] ci: register 8 framework / unit tests to run on AMD CI (#25939)

test/registered/disaggregation/test_specv2_kvcache_offloading.py

@@ -16,9 +16,10 @@ from sglang.srt.disaggregation.decode_kvcache_offload_manager import (
     DecodeKVCacheOffloadManager,
 )
 from sglang.srt.disaggregation.kv_events import OffloadedState
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=8, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=8, suite="stage-b-test-1-gpu-small-amd")
 
 
 def _make_mock_req(

test/registered/models/test_transformers_backend_eval.py

@@ -3,11 +3,16 @@
 import unittest
 from types import SimpleNamespace
 
-from sglang.test.ci.ci_register import register_cpu_ci, register_cuda_ci
+from sglang.test.ci.ci_register import (
+    register_amd_ci,
+    register_cpu_ci,
+    register_cuda_ci,
+)
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
 register_cuda_ci(est_time=48, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=48, suite="stage-b-test-1-gpu-small-amd")
 register_cpu_ci(est_time=320, suite="base-b-test-cpu")
 
 

test/registered/unit/constrained/test_e2e_constrained_reasoning.py

@@ -15,7 +15,7 @@ import unittest
 import requests
 
 from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
     CustomTestCase,
@@ -23,6 +23,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=120, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=120, suite="stage-b-test-1-gpu-small-amd")
 
 MODEL = "Qwen/Qwen3-0.6B"
 BASE_URL = "http://127.0.0.1:39877"

test/registered/unit/managers/test_priority_scheduling_disaggregation.py

@@ -12,9 +12,10 @@ from sglang.srt.disaggregation.decode import (  # noqa: E402
 from sglang.srt.disaggregation.utils import DisaggregationMode  # noqa: E402
 from sglang.srt.managers.schedule_batch import FINISH_ABORT  # noqa: E402
 from sglang.srt.managers.scheduler import Scheduler  # noqa: E402
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=5, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=5, suite="stage-b-test-1-gpu-small-amd")
 
 
 class TestDisaggregationPriorityQueueing(unittest.TestCase):

test/registered/unit/mem_cache/test_dsa_pool_host_unit.py

@@ -11,9 +11,10 @@ from sglang.srt.mem_cache.memory_pool_host import (
     alloc_with_pin_memory,
 )
 from sglang.srt.utils import is_cuda, is_hip, is_npu, is_xpu
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=9, suite="stage-b-test-1-gpu-small-amd")
 
 
 class TestDSAOffloadSignatures(unittest.TestCase):
@@ -143,6 +144,12 @@ class TestDSAHiCacheTransfer(unittest.TestCase):
                 ].cpu()
                 self.assertTrue(torch.equal(got_kv, expected_kv))
 
+    @unittest.skipIf(
+        is_hip(),
+        '`io_backend="kernel"` path in memory_pool_host.backup_from_device_all_layer '
+        "raises ValueError on AMD (only the `direct` IO backend is wired for ROCm). "
+        "The other 62 tests in this file pass on AMD.",
+    )
     def test_device_to_host_indexer_kernel(self):
         self._run_device_to_host_indexer_copy(io_backend="kernel")
 

test/registered/unit/mem_cache/test_swa_lock_release_lifecycle.py

@@ -29,10 +29,11 @@ from sglang.srt.mem_cache.radix_cache import RadixKey
 from sglang.srt.mem_cache.swa_memory_pool import SWAKVPool
 from sglang.srt.mem_cache.swa_radix_cache import SWARadixCache
 from sglang.srt.utils import get_device
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=12, suite="stage-b-test-1-gpu-small-amd")
 
 
 def _build_tree(

test/registered/unit/mem_cache/test_unified_radix_cache_bench.py

@@ -38,9 +38,10 @@ from sglang.srt.mem_cache.unified_cache_components.tree_component import Compone
 from sglang.srt.mem_cache.unified_radix_cache import UnifiedRadixCache
 from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
 from sglang.srt.utils import get_device
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=25, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=25, suite="stage-b-test-1-gpu-small-amd")
 
 # ---------------------------------------------------------------------------
 # Constants

test/registered/unit/mem_cache/test_unified_radix_cache_unittest.py

@@ -62,10 +62,11 @@ from sglang.srt.server_args import (
     set_global_server_args_for_scheduler,
 )
 from sglang.srt.utils import get_device
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=10, suite="stage-b-test-1-gpu-small-amd")
 
 
 @dataclass(frozen=True)