From 61bd0d490cbb148a6ef970673ac5d2d32e501bef Mon Sep 17 00:00:00 2001 From: FredricZ-2007 <226039983+fredricz-20070104@users.noreply.github.com> Date: Tue, 9 Sep 2025 08:47:16 +0000 Subject: [PATCH] feature: add rcca deepseekr1 int4 afp8 Signed-off-by: FredricZ-2007 <226039983+fredricz-20070104@users.noreply.github.com> --- tests/integration/defs/test_e2e.py | 38 ++++++++++++++++++- .../test_lists/qa/llm_function_core.txt | 1 + .../qa/llm_function_core_sanity.txt | 1 + 3 files changed, 38 insertions(+), 2 deletions(-) diff --git a/tests/integration/defs/test_e2e.py b/tests/integration/defs/test_e2e.py index 21bf49b363e3..840220a1225a 100644 --- a/tests/integration/defs/test_e2e.py +++ b/tests/integration/defs/test_e2e.py @@ -32,8 +32,8 @@ prune_checkpoint, quantize_data, refit_model, venv_check_call) from .conftest import (llm_models_root, skip_no_sm120, skip_nvlink_inactive, - skip_post_blackwell, skip_pre_blackwell, skip_pre_hopper, - tests_path, unittest_path) + skip_post_blackwell, skip_pre_ada, skip_pre_blackwell, + skip_pre_hopper, tests_path, unittest_path) sys.path.append(os.path.join(str(tests_path()), '/../examples/apps')) @@ -2155,6 +2155,40 @@ def test_relaxed_acceptance_quickstart_advanced_deepseek_r1_8gpus( _check_mem_usage(running_log, [85.6, 0, 0, 0], 8) +@skip_pre_ada +@skip_post_blackwell +@pytest.mark.skip_less_device_memory(80000) +@pytest.mark.skip_less_device(8) +@pytest.mark.parametrize("model_name,model_path", [ + pytest.param('DeepSeek-R1-W4AFP8', + 'DeepSeek-R1/DeepSeek-R1-W4AFP8', + marks=skip_pre_hopper), +]) +def test_ptp_quickstart_advanced_deepseek_r1_w4afp8_8gpus( + llm_root, llm_venv, model_name, model_path): + print(f"Testing {model_name}.") + example_root = Path(os.path.join(llm_root, "examples", "llm-api")) + with tempfile.NamedTemporaryFile(mode='w+t', + suffix=f".{model_name}.log", + dir="./", + delete=True, + delete_on_close=True) as running_log: + llm_venv.run_cmd([ + str(example_root / "quickstart_advanced.py"), + "--model_dir", + f"{llm_models_root()}/{model_path}", + "--moe_tp_size=1", + "--moe_ep_size=8", + "--tp_size=8", + "--use_cuda_graph", + f"--kv_cache_fraction={_MEM_FRACTION_50}", + "--max_batch_size=1", + "--max_seq_len=512", + ], + stdout=running_log) + _check_mem_usage(running_log, [50.0, 0, 0, 0], 8) + + @pytest.mark.skip_less_device_memory(80000) @pytest.mark.skip_less_device(8) @pytest.mark.parametrize("model_name,model_path", [ diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index 052484d7ddec..e2c3d53fb52f 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -645,6 +645,7 @@ test_e2e.py::test_ptp_quickstart_bert[TRTLLM-BertForSequenceClassification-bert/ test_e2e.py::test_ptp_star_attention_example[Llama3.1-8B-BF16-llama-3.1-model/Meta-Llama-3.1-8B] test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-hf-nvfp4-False-False] test_e2e.py::test_ptp_scaffolding[DeepSeek-R1-Distill-Qwen-7B-DeepSeek-R1/DeepSeek-R1-Distill-Qwen-7B] +test_e2e.py::test_ptp_quickstart_advanced_deepseek_r1_w4afp8_8gpus[DeepSeek-R1-W4AFP8-DeepSeek-R1/DeepSeek-R1-W4AFP8] unittest/llmapi/test_llm_pytorch.py::test_gemma3_1b_instruct_multi_lora examples/test_medusa.py::test_codellama_medusa_1gpu[CodeLlama-7b-Instruct] diff --git a/tests/integration/test_lists/qa/llm_function_core_sanity.txt b/tests/integration/test_lists/qa/llm_function_core_sanity.txt index 1fbb1277afc8..4df3e6797991 100644 --- a/tests/integration/test_lists/qa/llm_function_core_sanity.txt +++ b/tests/integration/test_lists/qa/llm_function_core_sanity.txt @@ -215,6 +215,7 @@ test_e2e.py::test_ptp_quickstart_multimodal[qwen2.5-vl-7b-instruct-Qwen2.5-VL-7B test_e2e.py::test_ptp_quickstart_multimodal[qwen2.5-vl-7b-instruct-Qwen2.5-VL-7B-Instruct-video-False] test_e2e.py::test_ptp_quickstart_multimodal[qwen2.5-vl-7b-instruct-Qwen2.5-VL-7B-Instruct-video-True] test_e2e.py::test_ptp_scaffolding[DeepSeek-R1-Distill-Qwen-7B-DeepSeek-R1/DeepSeek-R1-Distill-Qwen-7B] +test_e2e.py::test_ptp_quickstart_advanced_deepseek_r1_w4afp8_8gpus[DeepSeek-R1-W4AFP8-DeepSeek-R1/DeepSeek-R1-W4AFP8] test_e2e.py::test_ptp_star_attention_example[Llama3.1-8B-BF16-llama-3.1-model/Meta-Llama-3.1-8B] test_e2e.py::test_qwen_e2e_cpprunner_large_new_tokens[DeepSeek-R1-Distill-Qwen-1.5B-DeepSeek-R1-Distill-Qwen-1.5B] test_e2e.py::test_relaxed_acceptance_quickstart_advanced_deepseek_r1_8gpus[DeepSeek-R1-DeepSeek-R1/DeepSeek-R1]