[Bugfix] DiffusionGemma: cast the self-conditioning soft embed to the buffer dtype (#57462)

Signed-off-by: Matt Mastracci <matthew@mastracci.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
This commit is contained in:
Matt Mastracci
2026-09-20 04:58:02 +00:00
committed by GitHub
co-authored by Claude Fable 5.1 Lucas Wilkinson
parent 92b40f5a14
commit 0c90804c0b
@@ -543,7 +543,7 @@ def _compiled_sample_step(
if tp_size > 1:
soft_embeds = torch.ops.vllm.all_reduce(soft_embeds, group_name=tp_group_name)
soft_embeds = soft_embeds * normalizer
sc_embeds[decode_slots] = soft_embeds * sc_keep
sc_embeds[decode_slots] = (soft_embeds * sc_keep).to(sc_embeds.dtype)
# Overwrite canvas with argmax for newly converged denoise requests
newly_converged = (converged & is_denoise).unsqueeze(1)