From 0b225fb7b22f8ae1f5fc8ee618640ae0983c76de Mon Sep 17 00:00:00 2001 From: Komal Kumar Teru <162363718+kkt-cohere@users.noreply.github.com> Date: Mon, 2 Feb 2026 02:43:35 +0530 Subject: [PATCH] [Misc] skip target model mm emb in draft proposal step when draft is text-only (#33437) Signed-off-by: kkt-cohere --- vllm/v1/worker/gpu_model_runner.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/v1/worker/gpu_model_runner.py b/vllm/v1/worker/gpu_model_runner.py index 61e166133..061ac8680 100644 --- 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,