From b30b023563df33cb36000ba84fa41904caa95af6 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Mon, 13 Jul 2026 16:12:35 -0700 Subject: [PATCH 1/6] [TRTLLM-13948][test] Add DeepSeek R1/V3.2/V3-Lite disaggregated accuracy tests Signed-off-by: Asfiya Baig --- .../accuracy/test_disaggregated_serving.py | 132 ++++++++++++++++-- .../test_lists/qa/llm_function_core.txt | 7 +- .../test_lists/test-db/l0_dgx_b300.yml | 7 +- 3 files changed, 135 insertions(+), 11 deletions(-) diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index 59f4bcaa87f9..d460255a7d7c 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1350,13 +1350,15 @@ def test_guided_decoding(self, backend: str, mtp_nextn: int, mocker): @pytest.mark.skip_less_device(2) @pytest.mark.skip_less_device_memory(60000) @skip_pre_hopper - def test_kv_cache_v2_nixl_python(self): - """Test with use_kv_cache_manager_v2=True, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" + @pytest.mark.parametrize("use_kv_cache_manager_v2", [False, True], + ids=["cache_mgr_v1", "cache_mgr_v2"]) + def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): + """Test with KV cache manager v1 and v2, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" ctx_server_config = { "disable_overlap_scheduler": True, "kv_cache_config": { "enable_block_reuse": False, - "use_kv_cache_manager_v2": True + "use_kv_cache_manager_v2": use_kv_cache_manager_v2 }, "cache_transceiver_config": { "backend": "NIXL", @@ -1367,7 +1369,7 @@ def test_kv_cache_v2_nixl_python(self): "disable_overlap_scheduler": True, "kv_cache_config": { "enable_block_reuse": False, - "use_kv_cache_manager_v2": True + "use_kv_cache_manager_v2": use_kv_cache_manager_v2 }, "cache_transceiver_config": { "backend": "NIXL", @@ -1376,15 +1378,12 @@ def test_kv_cache_v2_nixl_python(self): } disaggregated_server_config = { "hostname": "localhost", - "port": 8000, "backend": "pytorch", "context_servers": { - "num_instances": 1, - "urls": ["localhost:8001"] + "num_instances": 1 }, "generation_servers": { - "num_instances": 1, - "urls": ["localhost:8002"] + "num_instances": 1 } } with launch_disaggregated_llm(disaggregated_server_config, @@ -1752,6 +1751,60 @@ def test_auto_dtype(self, overlap_scheduler): model_name=self.MODEL_NAME, test_sets=["MMLU", "GSM8K"]) + @pytest.mark.skip_less_device(4) + @pytest.mark.parametrize("use_kv_cache_manager_v2", [False, True], + ids=["cache_mgr_v1", "cache_mgr_v2"]) + def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): + """Test with KV cache manager v1 and v2, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" + max_num_tokens = 8192 + moe_config = {"backend": "TRTLLM", "max_num_tokens": max_num_tokens} + ctx_server_config = { + "disable_overlap_scheduler": True, + "kv_cache_config": { + "free_gpu_memory_fraction": 0.5, + "enable_block_reuse": False, + "use_kv_cache_manager_v2": use_kv_cache_manager_v2 + }, + "cache_transceiver_config": { + "backend": "NIXL", + "transceiver_runtime": "PYTHON", + "max_tokens_in_buffer": 4096 + }, + "tensor_parallel_size": 2, + "moe_expert_parallel_size": 2, + "enable_autotuner": False, + } + gen_server_config = { + "disable_overlap_scheduler": False, + "moe_config": moe_config, + "kv_cache_config": { + "free_gpu_memory_fraction": 0.5, + "enable_block_reuse": False, + "use_kv_cache_manager_v2": use_kv_cache_manager_v2 + }, + "cache_transceiver_config": { + "backend": "NIXL", + "transceiver_runtime": "PYTHON", + "max_tokens_in_buffer": 4096 + }, + "tensor_parallel_size": 2, + "moe_expert_parallel_size": 2, + "enable_autotuner": False, + } + disaggregated_server_config = { + "hostname": "localhost", + "backend": "pytorch", + "context_servers": { + "num_instances": 1, + }, + "generation_servers": { + "num_instances": 1, + } + } + with launch_disaggregated_llm(disaggregated_server_config, + ctx_server_config, gen_server_config, + self.MODEL_PATH) as llm: + run_accuracy_test(llm, self.MODEL_NAME, ["GSM8K"]) @pytest.mark.timeout(DEFAULT_TEST_TIMEOUT) @skip_pre_hopper @@ -2605,3 +2658,64 @@ def test_auto_dtype(self): server_waiting_timeout=3600) as llm: task = MMLU(self.MODEL_NAME) task.evaluate(llm, is_integration_test=True) + + +@pytest.mark.timeout(DEFAULT_TEST_TIMEOUT) +@skip_pre_blackwell +@pytest.mark.skip_less_device_memory(80000) +class TestDeepSeekR1(LlmapiAccuracyTestHarness): + MODEL_NAME = "deepseek-ai/DeepSeek-R1" + MODEL_PATH = f"{llm_models_root()}/DeepSeek-R1-0528-FP4-v2" + + @pytest.mark.skip_less_device(4) + @pytest.mark.parametrize("use_kv_cache_manager_v2", [False, True], + ids=["cache_mgr_v1", "cache_mgr_v2"]) + def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): + """Test with KV cache manager v1 and v2, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" + max_num_tokens = 8192 + moe_config = {"backend": "TRTLLM", "max_num_tokens": max_num_tokens} + ctx_server_config = { + "disable_overlap_scheduler": True, + "kv_cache_config": { + "free_gpu_memory_fraction": 0.5, + "enable_block_reuse": False, + "use_kv_cache_manager_v2": use_kv_cache_manager_v2 + }, + "cache_transceiver_config": { + "backend": "NIXL", + "transceiver_runtime": "PYTHON", + "max_tokens_in_buffer": 4096 + }, + "tensor_parallel_size": 2, + "moe_expert_parallel_size": 2, + } + gen_server_config = { + "disable_overlap_scheduler": False, + "moe_config": moe_config, + "kv_cache_config": { + "free_gpu_memory_fraction": 0.5, + "enable_block_reuse": False, + "use_kv_cache_manager_v2": use_kv_cache_manager_v2 + }, + "cache_transceiver_config": { + "backend": "NIXL", + "transceiver_runtime": "PYTHON", + "max_tokens_in_buffer": 4096 + }, + "tensor_parallel_size": 2, + "moe_expert_parallel_size": 2, + } + disaggregated_server_config = { + "hostname": "localhost", + "backend": "pytorch", + "context_servers": { + "num_instances": 1, + }, + "generation_servers": { + "num_instances": 1, + } + } + with launch_disaggregated_llm(disaggregated_server_config, + ctx_server_config, gen_server_config, + self.MODEL_PATH) as llm: + run_accuracy_test(llm, self.MODEL_NAME, ["GSM8K"]) diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index 02c43b89f8dd..808dfa75dfc7 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -1,4 +1,8 @@ +accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] +accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v2] accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] +accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1] +accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] @@ -10,7 +14,8 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python[cache_mgr_v1] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python[cache_mgr_v2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_nixl_backend accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[False] accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[True] diff --git a/tests/integration/test_lists/test-db/l0_dgx_b300.yml b/tests/integration/test_lists/test-db/l0_dgx_b300.yml index 5b53885685f9..4e2db110f307 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b300.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b300.yml @@ -64,8 +64,13 @@ l0_dgx_b300: - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto] - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8] - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend + - accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] + - accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v2] + - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1] + - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v2] - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_nixl_backend - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python[cache_mgr_v1] + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python[cache_mgr_v2] - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v1] - accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v2] From 2c93e52ed57c362c33551d44d303284ad8c8afab Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Tue, 14 Jul 2026 10:56:10 -0700 Subject: [PATCH 2/6] changes: 1. Revert changes to DeepSeekV3-Lite test 2. Update DeepSeekR1 and DeepSeekV3.2 tests to test only cache_mgr_v1 Signed-off-by: Asfiya Baig --- .../accuracy/test_disaggregated_serving.py | 22 +++++++++---------- .../test_lists/qa/llm_function_core.txt | 5 +---- .../test_lists/test-db/l0_dgx_b300.yml | 5 +---- 3 files changed, 12 insertions(+), 20 deletions(-) diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index d460255a7d7c..0f61136adccb 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1350,15 +1350,13 @@ def test_guided_decoding(self, backend: str, mtp_nextn: int, mocker): @pytest.mark.skip_less_device(2) @pytest.mark.skip_less_device_memory(60000) @skip_pre_hopper - @pytest.mark.parametrize("use_kv_cache_manager_v2", [False, True], - ids=["cache_mgr_v1", "cache_mgr_v2"]) - def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): - """Test with KV cache manager v1 and v2, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" + def test_kv_cache_v2_nixl_python(self): + """Test with use_kv_cache_manager_v2=True, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" ctx_server_config = { "disable_overlap_scheduler": True, "kv_cache_config": { "enable_block_reuse": False, - "use_kv_cache_manager_v2": use_kv_cache_manager_v2 + "use_kv_cache_manager_v2": True }, "cache_transceiver_config": { "backend": "NIXL", @@ -1369,7 +1367,7 @@ def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): "disable_overlap_scheduler": True, "kv_cache_config": { "enable_block_reuse": False, - "use_kv_cache_manager_v2": use_kv_cache_manager_v2 + "use_kv_cache_manager_v2": True }, "cache_transceiver_config": { "backend": "NIXL", @@ -1752,10 +1750,10 @@ def test_auto_dtype(self, overlap_scheduler): test_sets=["MMLU", "GSM8K"]) @pytest.mark.skip_less_device(4) - @pytest.mark.parametrize("use_kv_cache_manager_v2", [False, True], - ids=["cache_mgr_v1", "cache_mgr_v2"]) + @pytest.mark.parametrize("use_kv_cache_manager_v2", [False], + ids=["cache_mgr_v1"]) def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): - """Test with KV cache manager v1 and v2, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" + """Test with KV cache manager v1, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" max_num_tokens = 8192 moe_config = {"backend": "TRTLLM", "max_num_tokens": max_num_tokens} ctx_server_config = { @@ -2668,10 +2666,10 @@ class TestDeepSeekR1(LlmapiAccuracyTestHarness): MODEL_PATH = f"{llm_models_root()}/DeepSeek-R1-0528-FP4-v2" @pytest.mark.skip_less_device(4) - @pytest.mark.parametrize("use_kv_cache_manager_v2", [False, True], - ids=["cache_mgr_v1", "cache_mgr_v2"]) + @pytest.mark.parametrize("use_kv_cache_manager_v2", [False], + ids=["cache_mgr_v1"]) def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): - """Test with KV cache manager v1 and v2, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" + """Test with KV cache manager v1, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" max_num_tokens = 8192 moe_config = {"backend": "TRTLLM", "max_num_tokens": max_num_tokens} ctx_server_config = { diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index 808dfa75dfc7..8990687cc955 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -1,8 +1,6 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] -accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v2] accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1] -accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] @@ -14,8 +12,7 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python[cache_mgr_v1] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python[cache_mgr_v2] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_nixl_backend accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[False] accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[True] diff --git a/tests/integration/test_lists/test-db/l0_dgx_b300.yml b/tests/integration/test_lists/test-db/l0_dgx_b300.yml index 4e2db110f307..64f4d5632310 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b300.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b300.yml @@ -65,12 +65,9 @@ l0_dgx_b300: - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8] - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend - accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] - - accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v2] - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1] - - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v2] - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_nixl_backend - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python[cache_mgr_v1] - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python[cache_mgr_v2] + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v1] - accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v2] From 3749952f9bce8e4a00f47363ee670e9cbd49b7fc Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Tue, 14 Jul 2026 18:31:04 -0700 Subject: [PATCH 3/6] disable autotuner for DeepSeekR1 test Signed-off-by: Asfiya Baig --- tests/integration/defs/accuracy/test_disaggregated_serving.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index 0f61136adccb..99077e08360a 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -2686,6 +2686,7 @@ def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): }, "tensor_parallel_size": 2, "moe_expert_parallel_size": 2, + "enable_autotuner": False, } gen_server_config = { "disable_overlap_scheduler": False, @@ -2702,6 +2703,7 @@ def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): }, "tensor_parallel_size": 2, "moe_expert_parallel_size": 2, + "enable_autotuner": False, } disaggregated_server_config = { "hostname": "localhost", From 2f3cd02185c69394abb570604917188cf8ccf7d4 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Sun, 19 Jul 2026 14:09:49 -0700 Subject: [PATCH 4/6] address comments Signed-off-by: Asfiya Baig --- tests/integration/defs/accuracy/test_disaggregated_serving.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index 99077e08360a..1c1cdddb36d7 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1750,6 +1750,7 @@ def test_auto_dtype(self, overlap_scheduler): test_sets=["MMLU", "GSM8K"]) @pytest.mark.skip_less_device(4) + @pytest.mark.skip_less_device_memory(80000) @pytest.mark.parametrize("use_kv_cache_manager_v2", [False], ids=["cache_mgr_v1"]) def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): @@ -2663,7 +2664,7 @@ def test_auto_dtype(self): @pytest.mark.skip_less_device_memory(80000) class TestDeepSeekR1(LlmapiAccuracyTestHarness): MODEL_NAME = "deepseek-ai/DeepSeek-R1" - MODEL_PATH = f"{llm_models_root()}/DeepSeek-R1-0528-FP4-v2" + MODEL_PATH = f"{llm_models_root()}/DeepSeek-R1/DeepSeek-R1-0528-FP4-v2" @pytest.mark.skip_less_device(4) @pytest.mark.parametrize("use_kv_cache_manager_v2", [False], From 56f434ecf289b8a67be453f738345c28758abb8b Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Mon, 20 Jul 2026 11:23:55 -0700 Subject: [PATCH 5/6] pre-commit run formatting Signed-off-by: Asfiya Baig --- tests/integration/defs/accuracy/test_disaggregated_serving.py | 1 + 1 file changed, 1 insertion(+) diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index 1c1cdddb36d7..7b8471d11023 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1805,6 +1805,7 @@ def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): self.MODEL_PATH) as llm: run_accuracy_test(llm, self.MODEL_NAME, ["GSM8K"]) + @pytest.mark.timeout(DEFAULT_TEST_TIMEOUT) @skip_pre_hopper class TestQwen3_8B(LlmapiAccuracyTestHarness): From 762de1a553f095d9d32ea49e483042b07d150dc9 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Thu, 23 Jul 2026 13:56:25 -0700 Subject: [PATCH 6/6] Update memory guard to 200000 Signed-off-by: Asfiya Baig --- tests/integration/defs/accuracy/test_disaggregated_serving.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index 7b8471d11023..46d4f51a712c 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1750,7 +1750,7 @@ def test_auto_dtype(self, overlap_scheduler): test_sets=["MMLU", "GSM8K"]) @pytest.mark.skip_less_device(4) - @pytest.mark.skip_less_device_memory(80000) + @pytest.mark.skip_less_device_memory(200000) @pytest.mark.parametrize("use_kv_cache_manager_v2", [False], ids=["cache_mgr_v1"]) def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): @@ -2662,7 +2662,7 @@ def test_auto_dtype(self): @pytest.mark.timeout(DEFAULT_TEST_TIMEOUT) @skip_pre_blackwell -@pytest.mark.skip_less_device_memory(80000) +@pytest.mark.skip_less_device_memory(200000) class TestDeepSeekR1(LlmapiAccuracyTestHarness): MODEL_NAME = "deepseek-ai/DeepSeek-R1" MODEL_PATH = f"{llm_models_root()}/DeepSeek-R1/DeepSeek-R1-0528-FP4-v2"