diff --git a/.github/workflows/pr-test-rust.yml b/.github/workflows/pr-test-rust.yml index 3608bdaa2..f95cea28e 100644 --- a/.github/workflows/pr-test-rust.yml +++ b/.github/workflows/pr-test-rust.yml @@ -107,7 +107,7 @@ jobs: - name: Install SGLang dependencies run: | - sudo bash scripts/ci/ci_install_dependency.sh + sudo --preserve-env=PATH bash scripts/ci/ci_install_dependency.sh - name: Build python binding run: | diff --git a/sgl-router/README.md b/sgl-router/README.md index 72f94f034..ead374e4a 100644 --- a/sgl-router/README.md +++ b/sgl-router/README.md @@ -116,6 +116,7 @@ python -m sglang_router.launch_router \ ## Configuration + ### Logging Enable structured logging with optional file output: diff --git a/sgl-router/py_test/e2e/conftest.py b/sgl-router/py_test/e2e/conftest.py index c959095e7..460195816 100644 --- a/sgl-router/py_test/e2e/conftest.py +++ b/sgl-router/py_test/e2e/conftest.py @@ -692,7 +692,7 @@ def pytest_configure(config): @pytest.fixture(scope="session") def e2e_model() -> str: # Always use the default test model - return DEFAULT_MODEL_NAME_FOR_TEST + return os.getenv("E2E_PRIMARY_MODEL", DEFAULT_MODEL_NAME_FOR_TEST) @pytest.fixture diff --git a/sgl-router/py_test/e2e/test_pd_router.py b/sgl-router/py_test/e2e/test_pd_router.py index 93d92ad2a..c0ca06c3c 100644 --- a/sgl-router/py_test/e2e/test_pd_router.py +++ b/sgl-router/py_test/e2e/test_pd_router.py @@ -253,10 +253,10 @@ def test_pd_genai_bench(e2e_model: str, pd_cluster, genai_bench_runner): model_path=e2e_model, experiment_folder=policy_label, thresholds={ - "ttft_mean_max": 12, - "e2e_latency_mean_max": 15, - "input_throughput_mean_min": 400, - "output_throughput_mean_min": 20, + "ttft_mean_max": 13, + "e2e_latency_mean_max": 16, + "input_throughput_mean_min": 350, + "output_throughput_mean_min": 18, "gpu_util_p50_min": 99, }, kill_procs=pd_cluster.workers,