[Frontend] New allowed_token_ids decoding request parameter (#6753)

tests/entrypoints/openai/test_completion.py

@@ -541,6 +541,28 @@ async def test_logits_bias(client: openai.AsyncOpenAI):
     assert first_response != completion.choices[0].text
 
 
+@pytest.mark.asyncio
+async def test_allowed_token_ids(client: openai.AsyncOpenAI):
+    prompt = "Hello, my name is"
+    max_tokens = 1
+    tokenizer = get_tokenizer(tokenizer_name=MODEL_NAME)
+
+    # Test exclusive selection
+    allowed_ids = [21555, 21557, 21558]
+    completion = await client.completions.create(
+        model=MODEL_NAME,
+        prompt=prompt,
+        max_tokens=max_tokens,
+        temperature=0.0,
+        seed=42,
+        extra_body=dict(allowed_token_ids=allowed_ids),
+        logprobs=1,
+    )
+    response_tokens = completion.choices[0].logprobs.tokens
+    assert len(response_tokens) == 1
+    assert tokenizer.convert_tokens_to_ids(response_tokens)[0] in allowed_ids
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize("guided_decoding_backend",
                          ["outlines", "lm-format-enforcer"])