offload prompt_embeds decode in render_prompts_async to avoid blocking

vllm/renderers/base.py

@@ -104,6 +104,9 @@ def __init__(self, config: "VllmConfig", tokenizer: _T | None) -> None:
         self._process_multimodal_async = make_async(
             self._process_multimodal, executor=self._mm_executor
         )
+        self._safe_load_prompt_embeds_async = make_async(
+            safe_load_prompt_embeds, executor=self._executor
+        )
         if mm_registry.supports_multimodal_inputs(config.model_config):
             mm_processor_cache = mm_registry.processor_cache_from_config(config)
 
@@ -376,11 +379,28 @@ def render_prompts(
 
         return [self.render_prompt(prompt) for prompt in prompts]
 
+    async def _render_prompt_async(
+        self,
+        prompt: DictPrompt | bytes,
+    ) -> DictPrompt:
+        if isinstance(prompt, bytes):
+            embeds = await self._safe_load_prompt_embeds_async(
+                self.model_config, prompt
+            )
+            return EmbedsPrompt(prompt_embeds=embeds)
+
+        return prompt
+
     async def render_prompts_async(
         self,
         prompts: Sequence[DictPrompt | bytes],
     ) -> list[DictPrompt]:
-        return self.render_prompts(prompts)
+        if len(prompts) == 0:
+            raise ValueError("You must pass at least one prompt")
+
+        return await asyncio.gather(
+            *(self._render_prompt_async(prompt) for prompt in prompts)
+        )
 
     @abstractmethod
     def render_messages(