[Bugfix] Reject pipeline parallelism for DiffusionGemma (#45828)
Signed-off-by: Guan-Ming (Wesley) Chiu <105915352+guan404ming@users.noreply.github.com>
Guan-Ming Chiu committed
Jul 27, 2026 at 22:37 UTC
ef9975d021448b99a5408e8c78a4c4f6b63443c7
1 file changed
-6
vllm/model_executor/models/diffusion_gemma.py
-6
@@ -61,7 +61,6 @@ from vllm.v1.worker.gpu.states import RequestState
61
62
from .interfaces import (
63
SupportsMultiModal,
64
- SupportsPP,
64
SupportsQuant,
65
)
66
@@ -143,7 +142,6 @@ class DiffusionGemmaForConditionalGeneration(
142
nn.Module,
143
SupportsMultiModal,
144
SupportsQuant,
146
- SupportsPP,
145
):
146
"""DiffusionGemma for vLLM.
147
@@ -266,10 +264,6 @@ class DiffusionGemmaForConditionalGeneration(
264
eps=getattr(text_config, "rms_norm_eps", 1e-6),
265
)
266
269
- self.make_empty_intermediate_tensors = (
270
- self.model.make_empty_intermediate_tensors
271
- )
272
-
267
def compute_self_conditioning(
268
self,
269
inputs_embeds: torch.Tensor,