[Misc][OpenAI] deprecate max_tokens in favor of new max_completion_tokens field for chat completion endpoint

benchmarks/backend_request_func.py

@@ -324,7 +324,7 @@ async def async_request_openai_chat_completions(
                 },
             ],
             "temperature": 0.0,
-            "max_tokens": request_func_input.output_len,
+            "max_completion_tokens": request_func_input.output_len,
             "stream": True,
             "ignore_eos": request_func_input.ignore_eos,
         }

docs/source/serving/run_on_sky.rst

@@ -109,7 +109,7 @@ SkyPilot can scale up the service to multiple service replicas with built-in aut
       messages:
         - role: user
           content: Hello! What is your name?
-    max_tokens: 1
+    max_completion_tokens: 1
 
 .. raw:: html
 
@@ -129,7 +129,7 @@ SkyPilot can scale up the service to multiple service replicas with built-in aut
         messages:
           - role: user
             content: Hello! What is your name?
-        max_tokens: 1
+        max_completion_tokens: 1
 
   resources:
     accelerators: {L4, A10g, A10, L40, A40, A100, A100-80GB} # We can use cheaper accelerators for 8B model.
@@ -255,7 +255,7 @@ This will scale the service up to when the QPS exceeds 2 for each replica.
         messages:
           - role: user
             content: Hello! What is your name?
-        max_tokens: 1
+        max_completion_tokens: 1
 
   resources:
     accelerators: {L4, A10g, A10, L40, A40, A100, A100-80GB} # We can use cheaper accelerators for 8B model.

examples/offline_inference_openai.md

@@ -35,8 +35,8 @@
 
  ```
  $ cat openai_example_batch.jsonl
-{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": "Hello world!"}],"max_tokens": 1000}}
-{"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are an unhelpful assistant."},{"role": "user", "content": "Hello world!"}],"max_tokens": 1000}}
+{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}
+{"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are an unhelpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}
  ```
 
  ### Step 2: Run the batch
@@ -94,8 +94,8 @@ To follow along with this example, you can download the example batch, or create
 
  ```
  $ cat openai_example_batch.jsonl
-{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": "Hello world!"}],"max_tokens": 1000}}
-{"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are an unhelpful assistant."},{"role": "user", "content": "Hello world!"}],"max_tokens": 1000}}
+{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}
+{"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are an unhelpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}
  ```
 
 Now upload your batch file to your S3 bucket.

examples/openai_api_client_for_multimodal.py

@@ -53,7 +53,7 @@ def run_text_only() -> None:
             "content": "What's the capital of France?"
         }],
         model=model,
-        max_tokens=64,
+        max_completion_tokens=64,
     )
 
     result = chat_completion.choices[0].message.content
@@ -83,7 +83,7 @@ def run_single_image() -> None:
             ],
         }],
         model=model,
-        max_tokens=64,
+        max_completion_tokens=64,
     )
 
     result = chat_completion_from_url.choices[0].message.content
@@ -109,7 +109,7 @@ def run_single_image() -> None:
             ],
         }],
         model=model,
-        max_tokens=64,
+        max_completion_tokens=64,
     )
 
     result = chat_completion_from_base64.choices[0].message.content
@@ -144,7 +144,7 @@ def run_multi_image() -> None:
             ],
         }],
         model=model,
-        max_tokens=64,
+        max_completion_tokens=64,
     )
 
     result = chat_completion_from_url.choices[0].message.content
@@ -175,7 +175,7 @@ def run_audio() -> None:
             ],
         }],
         model=model,
-        max_tokens=64,
+        max_completion_tokens=64,
     )
 
     result = chat_completion_from_url.choices[0].message.content
@@ -201,7 +201,7 @@ def run_audio() -> None:
             ],
         }],
         model=model,
-        max_tokens=64,
+        max_completion_tokens=64,
     )
 
     result = chat_completion_from_base64.choices[0].message.content

examples/openai_example_batch.jsonl

@@ -1,2 +1,2 @@
-{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": "Hello world!"}],"max_tokens": 1000}}
-{"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are an unhelpful assistant."},{"role": "user", "content": "Hello world!"}],"max_tokens": 1000}}
+{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}
+{"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are an unhelpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}

requirements-common.txt

@@ -10,7 +10,7 @@ protobuf # Required by LlamaTokenizer.
 fastapi >= 0.107.0, < 0.113.0; python_version < '3.9'
 fastapi >= 0.107.0, != 0.113.*, != 0.114.0; python_version >= '3.9'
 aiohttp
-openai >= 1.40.0 # Ensure modern openai package (ensure types module present)
+openai >= 1.45.0 # Ensure modern openai package (ensure types module present and max_completion_tokens field support)
 uvicorn[standard]
 pydantic >= 2.9  # Required for fastapi >= 0.113.0
 pillow  # Required for image processing

tests/entrypoints/openai/test_audio.py

@@ -68,11 +68,12 @@ async def test_single_chat_session_audio(client: openai.AsyncOpenAI,
     }]
 
     # test single completion
-    chat_completion = await client.chat.completions.create(model=model_name,
-                                                           messages=messages,
-                                                           max_tokens=10,
-                                                           logprobs=True,
-                                                           top_logprobs=5)
+    chat_completion = await client.chat.completions.create(
+        model=model_name,
+        messages=messages,
+        max_completion_tokens=10,
+        logprobs=True,
+        top_logprobs=5)
     assert len(chat_completion.choices) == 1
 
     choice = chat_completion.choices[0]
@@ -91,7 +92,7 @@ async def test_single_chat_session_audio(client: openai.AsyncOpenAI,
     chat_completion = await client.chat.completions.create(
         model=model_name,
         messages=messages,
-        max_tokens=10,
+        max_completion_tokens=10,
     )
     message = chat_completion.choices[0].message
     assert message.content is not None and len(message.content) >= 0
@@ -123,11 +124,12 @@ async def test_single_chat_session_audio_base64encoded(
     }]
 
     # test single completion
-    chat_completion = await client.chat.completions.create(model=model_name,
-                                                           messages=messages,
-                                                           max_tokens=10,
-                                                           logprobs=True,
-                                                           top_logprobs=5)
+    chat_completion = await client.chat.completions.create(
+        model=model_name,
+        messages=messages,
+        max_completion_tokens=10,
+        logprobs=True,
+        top_logprobs=5)
     assert len(chat_completion.choices) == 1
 
     choice = chat_completion.choices[0]
@@ -146,7 +148,7 @@ async def test_single_chat_session_audio_base64encoded(
     chat_completion = await client.chat.completions.create(
         model=model_name,
         messages=messages,
-        max_tokens=10,
+        max_completion_tokens=10,
     )
     message = chat_completion.choices[0].message
     assert message.content is not None and len(message.content) >= 0
@@ -178,7 +180,7 @@ async def test_chat_streaming_audio(client: openai.AsyncOpenAI,
     chat_completion = await client.chat.completions.create(
         model=model_name,
         messages=messages,
-        max_tokens=10,
+        max_completion_tokens=10,
         temperature=0.0,
     )
     output = chat_completion.choices[0].message.content
@@ -188,7 +190,7 @@ async def test_chat_streaming_audio(client: openai.AsyncOpenAI,
     stream = await client.chat.completions.create(
         model=model_name,
         messages=messages,
-        max_tokens=10,
+        max_completion_tokens=10,
         temperature=0.0,
         stream=True,
     )
@@ -242,7 +244,7 @@ async def test_multi_audio_input(client: openai.AsyncOpenAI, model_name: str,
         await client.chat.completions.create(
             model=model_name,
             messages=messages,
-            max_tokens=10,
+            max_completion_tokens=10,
             temperature=0.0,
         )