From e2e3e2b8340b014a597ffa7368a7f22d250041a2 Mon Sep 17 00:00:00 2001 From: SeongJun Lee Date: Thu, 15 Oct 2026 02:04:07 +0900 Subject: [PATCH] [CI] Use an ungated model in the DGX Spark XQA test (#61416) Signed-off-by: lesj0610 Co-authored-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com> --- tests/v1/attention/test_attention_backends.py | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/tests/v1/attention/test_attention_backends.py b/tests/v1/attention/test_attention_backends.py index dc17c20227..b537a17823 100644 --- a/tests/v1/attention/test_attention_backends.py +++ b/tests/v1/attention/test_attention_backends.py @@ -1714,7 +1714,11 @@ def test_flashinfer_xqa_nvfp4_decode_correctness( from vllm.v1.attention.backends import flashinfer as fi seq_lens, bs, d = [40, 300, 1027], 16, 256 - config = create_vllm_config(max_model_len=max(seq_lens), block_size=bs) + config = create_vllm_config( + model_name="Qwen/Qwen3-0.6B", + max_model_len=max(seq_lens), + block_size=bs, + ) # GQA 6 at head_dim 256, as Qwen3.8-27B at TP=2. config.model_config.model_arch_config = replace( config.model_config.model_arch_config,