[Core] Registry for processing model inputs (#5214)

Co-authored-by: ywang96 <ywang@roblox.com>
2024-06-28 20:09:56 +08:00
parent 0d0e3a42ac
commit 5cbe8d155c
26 changed files with 784 additions and 398 deletions
--- a/vllm/engine/async_llm_engine.py
+++ b/vllm/engine/async_llm_engine.py
@@ -278,9 +278,11 @@ class _AsyncLLMEngine(LLMEngine):
        else:
            prompt_token_ids = inputs["prompt_token_ids"]

-        return LLMInputs(prompt_token_ids=prompt_token_ids,
-                         prompt=inputs.get("prompt"),
-                         multi_modal_data=inputs.get("multi_modal_data"))
+        llm_inputs = LLMInputs(prompt_token_ids=prompt_token_ids,
+                               prompt=inputs.get("prompt"),
+                               multi_modal_data=inputs.get("multi_modal_data"))
+
+        return self.input_processor(llm_inputs)

    async def add_request_async(
        self,