Uh oh!
There was an error while loading. Please reload this page.
skip invalid test case for helios pipeline - #13218
Conversation
Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
kaixuanliu
commented
Mar 6, 2026
@DN6 pls help review, thx! |
kaixuanliu
commented
Mar 16, 2026
@yiyixuxu Hi, can you help review? Thx! |
yiyixuxu
commented
Mar 16, 2026
hey @kaixuanliu did I miss anything? |
Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
kaixuanliu
commented
Mar 17, 2026
@yiyixuxu Thx for the comment! Yes, I am afraid you are right. But for this pipeline, it has a lot of modules using mixed precision, such as: keep_in_fp32_module, FP32LayerNorm, L435-L437 and HeliosOutputNorm, hence I suggest to skip this test case like in WanVACE pipeline, and I have updated the skip reason. Do you think it is OK? |
Uh oh!
There was an error while loading. Please reload this page.
* skip invalid test case for helio pipeline Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com> * update skip reason Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com> --------- Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
In helios pipeline, the latents are always in FP32 format: L798, hence we should skip the test_save_load_float16 test case, rather than adjust the tolerance.