From a964e5e6c35e8f22bd7663dcf93d1c801421a029 Mon Sep 17 00:00:00 2001 From: Cyrus Leung Date: Sun, 5 Oct 2025 13:38:53 +0800 Subject: [PATCH] [Bugfix] Allow `--skip-tokenizer-init` with `echo and return_token_ids` (#26238) Signed-off-by: DarkLight1337 --- tests/entrypoints/openai/test_token_in_token_out.py | 2 +- vllm/entrypoints/openai/serving_completion.py | 3 ++- 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/tests/entrypoints/openai/test_token_in_token_out.py b/tests/entrypoints/openai/test_token_in_token_out.py index f84605690c539..ed003939c44be 100644 --- a/tests/entrypoints/openai/test_token_in_token_out.py +++ b/tests/entrypoints/openai/test_token_in_token_out.py @@ -54,7 +54,7 @@ async def test_token_in_token_out_and_logprobs(server): prompt=token_ids, max_tokens=20, temperature=0, - echo=False, + echo=True, extra_body={ "return_token_ids": True, }, diff --git a/vllm/entrypoints/openai/serving_completion.py b/vllm/entrypoints/openai/serving_completion.py index 6e4113e6cf1ed..d0756e42b7963 100644 --- a/vllm/entrypoints/openai/serving_completion.py +++ b/vllm/entrypoints/openai/serving_completion.py @@ -691,5 +691,6 @@ class OpenAIServingCompletion(OpenAIServing): truncate_prompt_tokens=request.truncate_prompt_tokens, add_special_tokens=request.add_special_tokens, cache_salt=request.cache_salt, - needs_detokenization=bool(request.echo), + needs_detokenization=bool(request.echo + and not request.return_token_ids), )