From 33ef1941e217a2126d745caec6c6130d6aec3b31 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Nicol=C3=B2=20Lucchesi?= Date: Wed, 22 Apr 2026 15:21:02 +0200 Subject: [PATCH] [Bugfix][CI] Fix `v1/kv_connector/unit/test_nixl_connector_hma.py::test_fewer_blocks_with_hma` (#40597) Signed-off-by: NickLucche --- tests/v1/kv_connector/unit/test_nixl_connector_hma.py | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) diff --git a/tests/v1/kv_connector/unit/test_nixl_connector_hma.py b/tests/v1/kv_connector/unit/test_nixl_connector_hma.py index 5b60901735..3f5a9b9cc0 100644 --- a/tests/v1/kv_connector/unit/test_nixl_connector_hma.py +++ b/tests/v1/kv_connector/unit/test_nixl_connector_hma.py @@ -2,9 +2,11 @@ # SPDX-FileCopyrightText: Copyright contributors to the vLLM project """Unit tests for NixlConnectorScheduler with HMA and Mamba N-1 prefill.""" +import gc from unittest.mock import patch import pytest +import torch from vllm import LLM, SamplingParams from vllm.config import KVTransferConfig @@ -196,12 +198,13 @@ def test_fewer_blocks_with_hma(monkeypatch, model_name, sw_size): llm_kwargs = { "model": model_name, "enforce_eager": True, - "gpu_memory_utilization": 0.47, + "gpu_memory_utilization": 0.3, "kv_transfer_config": kv_transfer_config, "max_model_len": 2048, + "max_num_seqs": 1, # NOTE: Make sure HMA is enabled "disable_hybrid_kv_cache_manager": False, - "max_num_batched_tokens": 1024, + "max_num_batched_tokens": 2048, "enable_prefix_caching": False, "block_size": block_size, } @@ -248,6 +251,8 @@ def test_fewer_blocks_with_hma(monkeypatch, model_name, sw_size): assert len(group_block_ids) == expected_num_remote_blocks def run_test_and_cleanup(): + gc.collect() + torch.accelerator.empty_cache() llm = LLM(**llm_kwargs) try: run_hma_test(llm)