From 865f1f7fc8b361c2cf3a1918efb9b5555a77530a Mon Sep 17 00:00:00 2001 From: Levi <54832289+Levi-JQ@users.noreply.github.com> Date: Tue, 2 Dec 2025 22:21:47 +0800 Subject: [PATCH] [Bugfix] Resolve the interface compatibility issue of get_input_embeddings in MM (#4638) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit ### What this PR does / why we need it? Resolve the interface compatibility issue of get_input_embeddings in MM， because the get_input_embeddings func of other model does not have the is_multimodal parameter --------- Signed-off-by: Levi-JQ Co-authored-by: Levi-JQ --- vllm_ascend/worker/model_runner_v1.py | 18 +++++++++++++----- 1 file changed, 13 insertions(+), 5 deletions(-) diff --git a/vllm_ascend/worker/model_runner_v1.py b/vllm_ascend/worker/model_runner_v1.py index b9f460d..6fcb93b 100644 --- a/vllm_ascend/worker/model_runner_v1.py +++ b/vllm_ascend/worker/model_runner_v1.py @@ -1394,11 +1394,19 @@ class NPUModelRunner(LoRAModelRunnerMixin): # embeddings), we always use embeddings (rather than token ids) # as input to the multimodal model, even when the input is text. input_ids = self.input_ids[:total_num_scheduled_tokens] - inputs_embeds = self.model.get_input_embeddings( - input_ids, - multimodal_embeddings=mm_embeds, - is_multimodal=is_mm_embed, - ) + model_type = self.vllm_config.model_config.hf_config.model_type + if model_type == "qwen2_5_vl": + inputs_embeds = self.model.get_input_embeddings( + input_ids, + multimodal_embeddings=mm_embeds, + is_multimodal=is_mm_embed, + ) + else: + if mm_embeds: + inputs_embeds = self.model.get_input_embeddings( + input_ids, mm_embeds) + else: + inputs_embeds = self.model.get_input_embeddings(input_ids) # TODO(woosuk): Avoid the copy. Optimize. self.inputs_embeds[:total_num_scheduled_tokens].copy_( inputs_embeds)