diff --git a/vllm/model_executor/models/diffusion_gemma.py b/vllm/model_executor/models/diffusion_gemma.py index 0c1cc8c55e..10e0d8a2c2 100644 --- a/vllm/model_executor/models/diffusion_gemma.py +++ b/vllm/model_executor/models/diffusion_gemma.py @@ -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)