From 34b9899c8f1361964fded8d331550828c9258909 Mon Sep 17 00:00:00 2001 From: "Kevin H. Luu" Date: Tue, 8 Sep 2026 04:39:01 -0700 Subject: [PATCH] [CI] Increase ColQwen3 pooling test memory budget on H200 MIG (#55878) Signed-off-by: khluu --- tests/models/multimodal/pooling/test_colqwen3.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/tests/models/multimodal/pooling/test_colqwen3.py b/tests/models/multimodal/pooling/test_colqwen3.py index 2faac7fbfb..a9857bf4bd 100644 --- a/tests/models/multimodal/pooling/test_colqwen3.py +++ b/tests/models/multimodal/pooling/test_colqwen3.py @@ -45,7 +45,9 @@ TEXT_DOCUMENTS = [ ] DTYPE = "half" -GPU_MEMORY_UTILIZATION = 0.7 +# The Tomoro model needs room for its vision encoder and a 4096-token KV cache +# on the 16 GiB devices used by the H200 MIG test lane. +GPU_MEMORY_UTILIZATION = 0.8 def _make_base64_image(