From 0fdb52843ba4194b85b51dd910914b6349fdc367 Mon Sep 17 00:00:00 2001 From: guanweim Date: Mon, 10 Aug 2026 18:39:28 +0000 Subject: [PATCH] test(train): bump RLVR seq-length integ timeout to 3h GRPO RLVR on a 9B model (qwen3-5-9b) at 8K sequence length is far more compute-heavy than the 1B llama-based RLVR tests that share the 1h timeout. The job submits and runs correctly but does not reach Completed within 3600s, so the test asserted 'InProgress' == 'Completed' and failed. Raise max_wait_time to 10800 (3h), matching the budget already used by the Nova RLVR test in this file for a large model. --- .../tests/integ/train/test_rlvr_trainer_integration.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/sagemaker-train/tests/integ/train/test_rlvr_trainer_integration.py b/sagemaker-train/tests/integ/train/test_rlvr_trainer_integration.py index a022b6846f..bf2b348ddd 100644 --- a/sagemaker-train/tests/integ/train/test_rlvr_trainer_integration.py +++ b/sagemaker-train/tests/integ/train/test_rlvr_trainer_integration.py @@ -376,7 +376,7 @@ def test_rlvr_trainer_lora_with_sequence_length(sagemaker_session): training_job = rlvr_trainer.train(wait=False) logger.info(f"Training job submitted: {training_job.training_job_arn}") - max_wait_time = 3600 + max_wait_time = 10800 # 3h — GRPO RLVR on a 9B model at 8K is compute-heavy poll_interval = 30 start_time = time.time()