[Misc] skip target model mm emb in draft proposal step when draft is text-only (#33437)

Signed-off-by: kkt-cohere <komal@cohere.com>
2026-02-02 02:43:35 +05:30
parent 46b4a02794
commit 0b225fb7b2
1 changed files with 1 additions and 1 deletions
--- a/vllm/v1/worker/gpu_model_runner.py
+++ b/vllm/v1/worker/gpu_model_runner.py
@@ -4070,7 +4070,7 @@ class GPUModelRunner(
                    else:
                        target_hidden_states = hidden_states[:total_num_tokens]

-            if self.supports_mm_inputs:
+            if self.supports_mm_inputs and self.drafter.supports_mm_inputs:
                mm_embed_inputs = self._gather_mm_embeddings(
                    scheduler_output,
                    shift_computed_tokens=1,