From b99fdbe1f896abd7f4b133685a37397c3427f7d5 Mon Sep 17 00:00:00 2001 From: Ting SUN Date: Mon, 12 Oct 2026 00:27:21 +0800 Subject: [PATCH] [Bugfix] Respect the configured offline chat template (#61061) Signed-off-by: Ting Sun Co-authored-by: Cyrus Leung --- tests/entrypoints/multimodal/llm/test_mm_processor_kwargs.py | 1 + vllm/entrypoints/common/offline.py | 5 ++++- 2 files changed, 5 insertions(+), 1 deletion(-) diff --git a/tests/entrypoints/multimodal/llm/test_mm_processor_kwargs.py b/tests/entrypoints/multimodal/llm/test_mm_processor_kwargs.py index 91d2b0aa43..a381e0fa3b 100644 --- a/tests/entrypoints/multimodal/llm/test_mm_processor_kwargs.py +++ b/tests/entrypoints/multimodal/llm/test_mm_processor_kwargs.py @@ -14,6 +14,7 @@ from vllm.renderers.chat_utils import ChatCompletionMessageParam def _make_mock_llm() -> LLM: llm = object.__new__(LLM) + llm.chat_template = None llm.model_config = SimpleNamespace( runner_type="generate", enable_prompt_embeds=False ) diff --git a/vllm/entrypoints/common/offline.py b/vllm/entrypoints/common/offline.py index 1bb0bedccb..a5af009f22 100644 --- a/vllm/entrypoints/common/offline.py +++ b/vllm/entrypoints/common/offline.py @@ -54,6 +54,7 @@ class OfflineInferenceMixin: renderer: BaseRenderer llm_engine: "LLMEngine" model_config: ModelConfig + chat_template: str | None def _resolve_mm_lora( self, @@ -180,7 +181,9 @@ class OfflineInferenceMixin: renderer = self.renderer chat_params = ChatParams( - chat_template=chat_template, + chat_template=( + self.chat_template if chat_template is None else chat_template + ), chat_template_content_format=chat_template_content_format, chat_template_kwargs=merge_kwargs( chat_template_kwargs,