Replace pad with cat for better performance (#11388)

Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
2025-10-10 12:03:17 +08:00
parent 70fbb3adf6
commit b5044fbf12
5 changed files with 5 additions and 5 deletions
--- a/python/sglang/srt/models/dots_vlm_vit.py
+++ b/python/sglang/srt/models/dots_vlm_vit.py
@@ -323,7 +323,7 @@ class DotsVisionTransformer(PreTrainedModel):
            dim=0,
            dtype=grid_thw.dtype if torch.jit.is_tracing() else torch.int32,
        )
-        cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
+        cu_seqlens = torch.cat([cu_seqlens.new_zeros(1), cu_seqlens])

        for blk in self.blocks:
            hidden_states = blk(