From 910abdbd0866b203df2eb4e5419cada905d3803d Mon Sep 17 00:00:00 2001
From: Chauncey <chaunceyjiang@gmail.com>
Date: Sat, 11 Oct 2025 00:41:17 +0800
Subject: [PATCH] [Bugfix] fixed top_logprobs: -1 does not appear to work as
 intended (#26470)

Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
---
 tests/entrypoints/openai/test_chat_echo.py | 15 +++++++++++++++
 vllm/entrypoints/openai/serving_chat.py    |  2 +-
 2 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/tests/entrypoints/openai/test_chat_echo.py b/tests/entrypoints/openai/test_chat_echo.py
index a9c9c8e3df..b3b8b70033 100644
--- a/tests/entrypoints/openai/test_chat_echo.py
+++ b/tests/entrypoints/openai/test_chat_echo.py
@@ -7,12 +7,23 @@ import openai  # use the official client for correctness check
 import pytest
 import pytest_asyncio
 
+from vllm.config import ModelConfig
+
 from ...utils import RemoteOpenAIServer
 
 # # any model with a chat template should work here
 MODEL_NAME = "Qwen/Qwen2-1.5B-Instruct"
 
 
+def get_vocab_size(model_name):
+    config = ModelConfig(
+        model=model_name,
+        seed=0,
+        dtype="float16",
+    )
+    return config.get_vocab_size()
+
+
 @pytest.fixture(scope="module")
 def server():
     args = [
@@ -107,6 +118,7 @@ async def test_top_logprobs(client: openai.AsyncOpenAI):
     completion = await client.chat.completions.create(
         model=MODEL_NAME,
         messages=messages,
+        max_tokens=1,
         extra_body={
             "top_logprobs": -1,
             "logprobs": "true",
@@ -115,3 +127,6 @@ async def test_top_logprobs(client: openai.AsyncOpenAI):
     assert completion.choices[0].logprobs is not None
     assert completion.choices[0].logprobs.content is not None
     assert len(completion.choices[0].logprobs.content) > 0
+    assert len(
+        completion.choices[0].logprobs.content[0].top_logprobs
+    ) == get_vocab_size(MODEL_NAME)
diff --git a/vllm/entrypoints/openai/serving_chat.py b/vllm/entrypoints/openai/serving_chat.py
index d531806857..51c618e9d5 100644
--- a/vllm/entrypoints/openai/serving_chat.py
+++ b/vllm/entrypoints/openai/serving_chat.py
@@ -1643,7 +1643,7 @@ class OpenAIServingChat(OpenAIServing):
                 bytes=list(token.encode("utf-8", errors="replace")),
             )
             for i, p in enumerate(logprobs.items())
-            if top_logprobs and i < top_logprobs
+            if (top_logprobs and i < top_logprobs or top_logprobs == -1)
         ]
 
     def _create_chat_logprobs(