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 3, 15:15 UTC
[Bugfix] Fix LM detection for Nemotron Parse (#42641)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
May 14, 2026 at 23:42 UTC
b8a25d0e1216421b6218ca60796d73ead574bb3d
2 files changed
+5
-1
vllm/model_executor/models/interfaces.py
+2
-1
@@ -207,7 +207,8 @@ class SupportsMultiModal(Protocol):
207
208
raise NotImplementedError(
209
f"No language model found in {type(self).__name__}! "
210
- "You should initialize it via `_mark_language_model`."
210
+ "You should initialize it via `_mark_language_model`, "
211
+ "and make sure `embed_input_ids` is implemented."
212
)
213
214
@contextmanager
vllm/model_executor/models/nemotron_parse.py
+3
@@ -281,6 +281,9 @@ class MBartDecoderNoPos(nn.Module):
281
self.layernorm_embedding = nn.LayerNorm(config.d_model)
282
self.layer_norm = nn.LayerNorm(config.d_model)
283
284
+ def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
285
+ return self.embed_tokens(input_ids)
286
+
287
def forward(
288
self,
289
decoder_input_ids: torch.Tensor | None,