[CI] Increase ColQwen3 pooling test memory budget on H200 MIG (#55878)
Signed-off-by: khluu <khluu000@gmail.com>
Kevin H. Luu committed
Sep 8, 2026 at 04:39 UTC
34b9899c8f1361964fded8d331550828c9258909
1 file changed
+3
-1
tests/models/multimodal/pooling/test_colqwen3.py
+3
-1
@@ -45,7 +45,9 @@ TEXT_DOCUMENTS = [
45
]
46
47
DTYPE = "half"
48
-GPU_MEMORY_UTILIZATION = 0.7
48
+# The Tomoro model needs room for its vision encoder and a 4096-token KV cache
49
+# on the 16 GiB devices used by the H200 MIG test lane.
50
+GPU_MEMORY_UTILIZATION = 0.8
51
52
53
def _make_base64_image(