Fix echo/logprob OpenAI completion bug (#3441)

Co-authored-by: Dylan Hawk <dylanwawk@gmail.com>

tests/entrypoints/test_openai_server.py

@@ -742,5 +742,36 @@ number: "1" | "2"
     assert content.strip() == ground_truth
 
 
+@pytest.mark.parametrize(
+    # first test base model, then test loras
+    "model_name",
+    [MODEL_NAME, "zephyr-lora", "zephyr-lora2"],
+)
+async def test_echo_logprob_completion(server, client: openai.AsyncOpenAI,
+                                       model_name: str):
+    tokenizer = get_tokenizer(tokenizer_name=MODEL_NAME)
+    # test using text and token IDs
+    for prompt in ("Hello, my name is", [0, 0, 0, 0, 0]):
+        completion = await client.completions.create(model=model_name,
+                                                     prompt=prompt,
+                                                     max_tokens=5,
+                                                     temperature=0.0,
+                                                     echo=True,
+                                                     logprobs=1)
+
+        prompt_text = tokenizer.decode(prompt) if isinstance(prompt,
+                                                             list) else prompt
+        assert (completion.choices[0].text is not None
+                and re.search(r"^" + prompt_text, completion.choices[0].text))
+        logprobs = completion.choices[0].logprobs
+        assert logprobs is not None
+        assert len(logprobs.text_offset) > 5
+        assert (len(logprobs.token_logprobs) > 5
+                and logprobs.token_logprobs[0] is None)
+        assert (len(logprobs.top_logprobs) > 5
+                and logprobs.top_logprobs[0] is None)
+        assert len(logprobs.tokens) > 5
+
+
 if __name__ == "__main__":
     pytest.main([__file__])