Unverified Commit d72b0be3 authored by Chendi.Xue's avatar Chendi.Xue Committed by GitHub
Browse files

[XPU]Fix for Qwen-OMNI crash (#35249)


Signed-off-by: default avatarChendi Xue <chendi.xue@intel.com>
parent 42489e43
...@@ -105,9 +105,10 @@ class xpu_ops: ...@@ -105,9 +105,10 @@ class xpu_ops:
assert len(window_size) == 2 assert len(window_size) == 2
real_window_size = (window_size[0], window_size[1]) # noqa: F841 real_window_size = (window_size[0], window_size[1]) # noqa: F841
# In encode attention, v maybe not contiguous and current # In encode attention, k and v maybe not contiguous and current
# kernel can't handle it # kernel can't handle it
if block_table is None: if block_table is None:
k = k.contiguous()
v = v.contiguous() v = v.contiguous()
return flash_attn_varlen_func( return flash_attn_varlen_func(
out=out, out=out,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment