[Bugfix] Respect the configured offline chat template (#61061)

Signed-off-by: Ting Sun <suntcrick@gmail.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>

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
     )

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,