A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 6, 13:15 UTC
[Model] Remove unused `handle_oov_mm_token` (#37321)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Mar 18, 2026 at 03:44 UTC
51f0acda7960871f9fdc81d79481b18bee957ea8
3 files changed
+4
-4
tests/models/multimodal/processing/test_qwen2_5_omni_embed.py
+4
-2
@@ -185,14 +185,16 @@ def make_mock_model(hidden: int = 8):
185
186
# super().embed_input_ids → use SupportsMultiModal.embed_input_ids
187
def fake_super_embed(
188
- ids, mm_embs=None, *, is_multimodal=None, handle_oov_mm_token=False
188
+ ids,
189
+ mm_embs=None,
190
+ *,
191
+ is_multimodal=None,
192
):
193
return SupportsMultiModal.embed_input_ids(
194
model,
195
ids,
196
mm_embs,
197
is_multimodal=is_multimodal,
195
- handle_oov_mm_token=handle_oov_mm_token,
198
)
199
200
# Bind embed_input_ids as the real method
vllm/model_executor/models/fireredasr2.py
-1
@@ -793,7 +793,6 @@ class FireRedASR2ForConditionalGeneration(
793
multimodal_embeddings: MultiModalEmbeddings | None = None,
794
*,
795
is_multimodal: torch.Tensor | None = None,
796
- handle_oov_mm_token: bool = False,
796
) -> torch.Tensor:
797
inputs_embeds = self.model.decoder.embed_input_ids(input_ids)
798
vllm/model_executor/models/kimi_audio.py
-1
@@ -514,7 +514,6 @@ class KimiAudioForConditionalGeneration(
514
multimodal_embeddings: tuple[torch.Tensor, ...] | None = None,
515
*,
516
is_multimodal: torch.Tensor | None = None,
517
- handle_oov_mm_token: bool = False,
517
) -> torch.Tensor:
518
"""Embed input IDs and fuse with audio embeddings.
519