[AMD] ci: register 5 framework tests to run on AMD CI (#25208)

Co-authored-by: HAI <hixiao@gmail.com>

test/registered/core/test_engine_child_pids.py

@@ -14,13 +14,14 @@ import unittest
 import psutil
 
 import sglang as sgl
-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_SMALL_MODEL_NAME_FOR_TEST,
     CustomTestCase,
 )
 
 register_cuda_ci(est_time=77, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=77, suite="stage-b-test-1-gpu-small-amd")
 
 
 class TestEngineChildPids(CustomTestCase):

test/registered/observability/test_tracing.py

@@ -34,7 +34,7 @@ from sglang.srt.observability.trace import (
 )
 from sglang.srt.utils import kill_process_tree
 from sglang.srt.utils.network import get_zmq_socket
-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_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -47,6 +47,7 @@ logger = logging.getLogger(__name__)
 
 # CI registration
 register_cuda_ci(est_time=113, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=113, suite="stage-b-test-1-gpu-small-amd")
 
 
 # ============================================================================

test/registered/prefill_only/test_pooled_hidden_states.py

@@ -17,8 +17,8 @@ import requests
 import torch
 
 from sglang.srt.entrypoints.engine import Engine
-from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.srt.utils import is_hip, kill_process_tree
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import (
     DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -28,6 +28,8 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=100, stage="base-b", runner_config="1-gpu-small")
+register_amd_ci(est_time=100, suite="stage-b-test-1-gpu-small-amd")
+
 
 _SEQCLS_MODEL = "Qwen/Qwen3-0.6B"
 _CAUSAL_LM_MODEL = DEFAULT_SMALL_MODEL_NAME_FOR_TEST
@@ -183,6 +185,11 @@ class TestPooledHiddenStatesEngine(CustomTestCase):
 # ---------------------------------------------------------------------------
 
 
+@unittest.skipIf(
+    is_hip(),
+    "Multi-Item Scoring (enable_mis) requires the flashinfer prefill/decode "
+    "backend, which is NVIDIA-only.",
+)
 class TestPooledHiddenStatesMISEngine(CustomTestCase):
     """Validates return_pooled_hidden_states in MIS (delimiter) scoring mode.
 

test/registered/sessions/test_session_control.py

@@ -13,9 +13,9 @@ import unittest
 import aiohttp
 import requests
 
-from sglang.srt.utils import kill_process_tree
+from sglang.srt.utils import is_hip, kill_process_tree
 from sglang.srt.utils.hf_transformers_utils import get_tokenizer
-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_SMALL_MODEL_NAME_FOR_TEST,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -25,6 +25,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=87, stage="extra-a", runner_config="1-gpu-large")
+register_amd_ci(est_time=87, suite="stage-b-test-1-gpu-large-amd")
 
 
 def remove_prefix(text: str, prefix: str) -> str:
@@ -550,6 +551,14 @@ class TestSessionControl(CustomTestCase):
             outputs_from_session == outputs_normal
         ), f"outputs_from_session: {outputs_from_session}, outputs_normal: {outputs_normal}"
 
+    @unittest.skipIf(
+        is_hip(),
+        "Session-branching produces a deterministic 1-token divergence from "
+        "plain generation on AMD (greedy temperature=0). The other 4 of 5 "
+        "branch outputs match exactly; passes on CUDA. Suspected ROCm "
+        "numerical-precision difference in the session KV-cache reuse path. "
+        "Re-enable when that divergence is fixed.",
+    )
     def test_session_control_with_branching(self):
         root_prompt = "First, let me explain in one sentence about AI"
         chunks_per_step = [

test/registered/sessions/test_streaming_session.py

@@ -7,7 +7,7 @@ test_streaming_session_extra.py.
 
 import unittest
 
-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.kits.streaming_session_kit import (
     AbortLeakReproKitMixin,
     StreamingSessionKitMixin,
@@ -24,6 +24,7 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=691, stage="base-b", runner_config="1-gpu-large")
+register_amd_ci(est_time=691, suite="stage-b-test-1-gpu-large-amd")
 
 
 class TestStreamingSession(StreamingSessionServerBase, StreamingSessionKitMixin):