AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-10-27 21:33:02 +03:00
Code Activity

[tests] skip hunyuan framepack group offloading tests (#14551)

Browse Source 
skip hunyuan framepack group offloading tests
This commit is contained in:
Sayak Paul
2026-08-22 13:13:24 +05:30
committed by GitHub
parent ecfbf7662c
commit 1b98ae1060
1 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/pipelines/hunyuan_video/test_hunyuan_video_framepack.py
View File

@@ -389,6 +389,14 @@ class HunyuanVideoFramepackPipelineFastTests(
# this test because of MHA (example: HunyuanDiT because of AttentionPooling layer).
pass
@unittest.skip("The image_encoder uses SiglipVisionModel, which does not support group offloading.")
def test_pipeline_level_group_offloading_inference(self):
# Same root cause as the sequential CPU offloading skips above: the attention layer is a
# torch.nn.MultiheadAttention, which passes `self.out_proj.weight` to
# `torch.nn.functional.multi_head_attention_forward` instead of calling `self.out_proj`. The leaf-level
# onload hook on `out_proj` is therefore never triggered and its weights stay on the CPU.
pass
# TODO(aryan): Create a dummy gemma model with smol vocab size
@unittest.skip(
"A very small vocab size is used for fast tests. So, Any kind of prompt other than the empty default used in other tests will lead to a embedding lookup error. This test uses a long prompt that causes the error."