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 9, 03:45 UTC
[Misc] skip target model mm emb in draft proposal step when draft is text-only (#33437)
Signed-off-by: kkt-cohere <komal@cohere.com>
Komal Kumar Teru committed
Feb 2, 2026 at 02:43 UTC
0b225fb7b22f8ae1f5fc8ee618640ae0983c76de
1 file changed
+1
-1
vllm/v1/worker/gpu_model_runner.py
+1
-1
@@ -4070,7 +4070,7 @@ class GPUModelRunner(
4070
else:
4071
target_hidden_states = hidden_states[:total_num_tokens]
4072
4073
- if self.supports_mm_inputs:
4073
+ if self.supports_mm_inputs and self.drafter.supports_mm_inputs:
4074
mm_embed_inputs = self._gather_mm_embeddings(
4075
scheduler_output,
4076
shift_computed_tokens=1,