diff --git a/sgl-model-gateway/e2e_test/infra/model_pool.py b/sgl-model-gateway/e2e_test/infra/model_pool.py index 98e4b5e9b..d928cc7c3 100644 --- a/sgl-model-gateway/e2e_test/infra/model_pool.py +++ b/sgl-model-gateway/e2e_test/infra/model_pool.py @@ -466,7 +466,7 @@ class ModelPool: "[%.1fs] %s is healthy at %s (check #%d)", elapsed, key, - instance.base_url, + instance.worker_url, check_count, ) instance._healthy = True diff --git a/sgl-model-gateway/e2e_test/infra/run_eval.py b/sgl-model-gateway/e2e_test/infra/run_eval.py index 672c885b5..753e27df9 100644 --- a/sgl-model-gateway/e2e_test/infra/run_eval.py +++ b/sgl-model-gateway/e2e_test/infra/run_eval.py @@ -85,6 +85,7 @@ def run_eval(args: Any) -> dict: # Build base URL base_url = getattr(args, "base_url", None) if base_url: + base_url = base_url.rstrip("/") # Remove trailing slashes if not base_url.endswith("/v1"): base_url = f"{base_url}/v1" else: diff --git a/sgl-model-gateway/e2e_test/router/test_mmlu.py b/sgl-model-gateway/e2e_test/router/test_mmlu.py index a9a6ab11e..770211125 100644 --- a/sgl-model-gateway/e2e_test/router/test_mmlu.py +++ b/sgl-model-gateway/e2e_test/router/test_mmlu.py @@ -36,10 +36,9 @@ class TestMMLU: Note: setup_backend fixture already waits for workers to be ready. """ backend, model, client, *_ = setup_backend - base_url = str(client.base_url).rstrip("/v1") args = SimpleNamespace( - base_url=base_url, + base_url=str(client.base_url), model=model, eval_name="mmlu", num_examples=64, @@ -60,10 +59,9 @@ class TestMMLU: significant results. """ backend, model, client, *_ = setup_backend - base_url = str(client.base_url).rstrip("/v1") args = SimpleNamespace( - base_url=base_url, + base_url=str(client.base_url), model=model, eval_name="mmlu", num_examples=128, diff --git a/sgl-model-gateway/e2e_test/router/test_pd_mmlu.py b/sgl-model-gateway/e2e_test/router/test_pd_mmlu.py index e6b740bbd..bd2a4ae8f 100644 --- a/sgl-model-gateway/e2e_test/router/test_pd_mmlu.py +++ b/sgl-model-gateway/e2e_test/router/test_pd_mmlu.py @@ -44,10 +44,9 @@ class TestPDMMLU: accuracy meets threshold (>= 0.65). """ backend, model, client, *_ = setup_backend - base_url = str(client.base_url).rstrip("/v1") args = SimpleNamespace( - base_url=base_url, + base_url=str(client.base_url), model=model, eval_name="mmlu", num_examples=64,