(feat) add Predicted Outputs for OpenAI (BerriAI#6594)

* bump openai to openai==1.54.0

* add 'prediction' param

* testing fix bedrock deprecated cohere.command-text-v14

* test test_openai_prediction_param.py

* test_openai_prediction_param_with_caching

* doc Predicted Outputs

* doc Predicted Output

.circleci/config.yml

@@ -47,7 +47,7 @@ jobs:
             pip install opentelemetry-api==1.25.0
             pip install opentelemetry-sdk==1.25.0
             pip install opentelemetry-exporter-otlp==1.25.0
-            pip install openai==1.52.0
+            pip install openai==1.54.0 
             pip install prisma==0.11.0   
             pip install "detect_secrets==1.5.0"         
             pip install "httpx==0.24.1"
@@ -520,7 +520,7 @@ jobs:
             pip install "aiodynamo==23.10.1"
             pip install "asyncio==3.4.3"
             pip install "PyGithub==1.59.1"
-            pip install "openai==1.52.0"
+            pip install "openai==1.54.0 "
             # Run pytest and generate JUnit XML report
       - run:
           name: Build Docker image
@@ -637,7 +637,7 @@ jobs:
             pip install "aiodynamo==23.10.1"
             pip install "asyncio==3.4.3"
             pip install "PyGithub==1.59.1"
-            pip install "openai==1.52.0"
+            pip install "openai==1.54.0 "
       - run:
           name: Build Docker image
           command: docker build -t my-app:latest -f ./docker/Dockerfile.database .
@@ -729,7 +729,7 @@ jobs:
             pip install "pytest-asyncio==0.21.1"
             pip install "google-cloud-aiplatform==1.43.0"
             pip install aiohttp
-            pip install "openai==1.52.0"
+            pip install "openai==1.54.0 "
             python -m pip install --upgrade pip
             pip install "pydantic==2.7.1"
             pip install "pytest==7.3.1"
@@ -924,7 +924,7 @@ jobs:
             pip install "pytest-retry==1.6.3"
             pip install "pytest-asyncio==0.21.1"
             pip install aiohttp
-            pip install "openai==1.52.0"
+            pip install "openai==1.54.0 "
             python -m pip install --upgrade pip
             pip install "pydantic==2.7.1"
             pip install "pytest==7.3.1"

.circleci/requirements.txt

@@ -1,5 +1,5 @@
 # used by CI/CD testing
-openai==1.52.0
+openai==1.54.0 
 python-dotenv
 tiktoken
 importlib_metadata

docs/my-website/docs/completion/predict_outputs.md

@@ -0,0 +1,109 @@
+import Tabs from '@theme/Tabs';
+import TabItem from '@theme/TabItem';
+
+# Predicted Outputs
+
+| Property | Details |
+|-------|-------|
+| Description | Use this when most of the output of the LLM is known ahead of time. For instance, if you are asking the model to rewrite some text or code with only minor changes, you can reduce your latency significantly by using Predicted Outputs, passing in the existing content as your prediction. |
+| Supported providers | `openai` |
+| Link to OpenAI doc on Predicted Outputs | [Predicted Outputs ↗](https://platform.openai.com/docs/guides/latency-optimization#use-predicted-outputs) |
+| Supported from LiteLLM Version | `v1.51.4` |
+
+
+
+## Using Predicted Outputs
+
+<Tabs>
+<TabItem label="LiteLLM Python SDK" value="Python">
+
+In this example we want to refactor a piece of C# code, and convert the Username property to Email instead:
+```python
+import litellm
+os.environ["OPENAI_API_KEY"] = "your-api-key"
+code = """
+/// <summary>
+/// Represents a user with a first name, last name, and username.
+/// </summary>
+public class User
+{
+    /// <summary>
+    /// Gets or sets the user's first name.
+    /// </summary>
+    public string FirstName { get; set; }
+
+    /// <summary>
+    /// Gets or sets the user's last name.
+    /// </summary>
+    public string LastName { get; set; }
+
+    /// <summary>
+    /// Gets or sets the user's username.
+    /// </summary>
+    public string Username { get; set; }
+}
+"""
+
+completion = litellm.completion(
+    model="gpt-4o-mini",
+    messages=[
+        {
+            "role": "user",
+            "content": "Replace the Username property with an Email property. Respond only with code, and with no markdown formatting.",
+        },
+        {"role": "user", "content": code},
+    ],
+    prediction={"type": "content", "content": code},
+)
+
+print(completion)
+```
+
+</TabItem>
+<TabItem label="LiteLLM Proxy Server" value="proxy">
+
+1. Define models on config.yaml
+
+```yaml
+model_list:
+  - model_name: gpt-4o-mini # OpenAI gpt-4o-mini
+    litellm_params:
+      model: openai/gpt-4o-mini
+      api_key: os.environ/OPENAI_API_KEY 
+
+```
+
+2. Run proxy server
+
+```bash
+litellm --config config.yaml
+```
+
+3. Test it using the OpenAI Python SDK
+
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    api_key="LITELLM_PROXY_KEY", # sk-1234
+    base_url="LITELLM_PROXY_BASE" # http://0.0.0.0:4000
+)
+
+completion = client.chat.completions.create(
+    model="gpt-4o-mini",
+    messages=[
+        {
+            "role": "user",
+            "content": "Replace the Username property with an Email property. Respond only with code, and with no markdown formatting.",
+        },
+        {"role": "user", "content": code},
+    ],
+    prediction={"type": "content", "content": code},
+)
+
+print(completion)
+```
+
+</TabItem>
+</Tabs>

docs/my-website/sidebars.js

@@ -205,6 +205,7 @@ const sidebars = {
         "completion/prompt_caching",
         "completion/audio",
         "completion/vision",
+        "completion/predict_outputs",
         "completion/prefix",
         "completion/drop_params",
         "completion/prompt_formatting",

litellm/llms/OpenAI/chat/gpt_transformation.py

@@ -94,6 +94,7 @@ def get_supported_openai_params(self, model: str) -> list:
             "max_tokens",
             "max_completion_tokens",
             "modalities",
+            "prediction",
             "n",
             "presence_penalty",
             "seed",

litellm/main.py

@@ -162,6 +162,7 @@
     ChatCompletionAssistantMessage,
     ChatCompletionAudioParam,
     ChatCompletionModality,
+    ChatCompletionPredictionContentParam,
     ChatCompletionUserMessage,
     HttpxBinaryResponseContent,
 )
@@ -304,6 +305,7 @@ async def acompletion(
     max_tokens: Optional[int] = None,
     max_completion_tokens: Optional[int] = None,
     modalities: Optional[List[ChatCompletionModality]] = None,
+    prediction: Optional[ChatCompletionPredictionContentParam] = None,
     audio: Optional[ChatCompletionAudioParam] = None,
     presence_penalty: Optional[float] = None,
     frequency_penalty: Optional[float] = None,
@@ -346,6 +348,7 @@ async def acompletion(
         max_tokens (integer, optional): The maximum number of tokens in the generated completion (default is infinity).
         max_completion_tokens (integer, optional): An upper bound for the number of tokens that can be generated for a completion, including visible output tokens and reasoning tokens.
         modalities (List[ChatCompletionModality], optional): Output types that you would like the model to generate for this request. You can use `["text", "audio"]`
+        prediction (ChatCompletionPredictionContentParam, optional): Configuration for a Predicted Output, which can greatly improve response times when large parts of the model response are known ahead of time. This is most common when you are regenerating a file with only minor changes to most of the content.
         audio (ChatCompletionAudioParam, optional): Parameters for audio output. Required when audio output is requested with modalities: ["audio"]
         presence_penalty (float, optional): It is used to penalize new tokens based on their existence in the text so far.
         frequency_penalty: It is used to penalize new tokens based on their frequency in the text so far.
@@ -387,6 +390,7 @@ async def acompletion(
         "max_tokens": max_tokens,
         "max_completion_tokens": max_completion_tokens,
         "modalities": modalities,
+        "prediction": prediction,
         "audio": audio,
         "presence_penalty": presence_penalty,
         "frequency_penalty": frequency_penalty,
@@ -693,6 +697,7 @@ def completion(  # type: ignore # noqa: PLR0915
     max_completion_tokens: Optional[int] = None,
     max_tokens: Optional[int] = None,
     modalities: Optional[List[ChatCompletionModality]] = None,
+    prediction: Optional[ChatCompletionPredictionContentParam] = None,
     audio: Optional[ChatCompletionAudioParam] = None,
     presence_penalty: Optional[float] = None,
     frequency_penalty: Optional[float] = None,
@@ -737,6 +742,7 @@ def completion(  # type: ignore # noqa: PLR0915
         max_tokens (integer, optional): The maximum number of tokens in the generated completion (default is infinity).
         max_completion_tokens (integer, optional): An upper bound for the number of tokens that can be generated for a completion, including visible output tokens and reasoning tokens.
         modalities (List[ChatCompletionModality], optional): Output types that you would like the model to generate for this request.. You can use `["text", "audio"]`
+        prediction (ChatCompletionPredictionContentParam, optional): Configuration for a Predicted Output, which can greatly improve response times when large parts of the model response are known ahead of time. This is most common when you are regenerating a file with only minor changes to most of the content.
         audio (ChatCompletionAudioParam, optional): Parameters for audio output. Required when audio output is requested with modalities: ["audio"]
         presence_penalty (float, optional): It is used to penalize new tokens based on their existence in the text so far.
         frequency_penalty: It is used to penalize new tokens based on their frequency in the text so far.
@@ -843,6 +849,7 @@ def completion(  # type: ignore # noqa: PLR0915
         "stop",
         "max_completion_tokens",
         "modalities",
+        "prediction",
         "audio",
         "max_tokens",
         "presence_penalty",
@@ -994,6 +1001,7 @@ def completion(  # type: ignore # noqa: PLR0915
             max_tokens=max_tokens,
             max_completion_tokens=max_completion_tokens,
             modalities=modalities,
+            prediction=prediction,
             audio=audio,
             presence_penalty=presence_penalty,
             frequency_penalty=frequency_penalty,

litellm/types/llms/openai.py

@@ -21,6 +21,9 @@
 from openai.types.chat import ChatCompletionChunk
 from openai.types.chat.chat_completion_audio_param import ChatCompletionAudioParam
 from openai.types.chat.chat_completion_modality import ChatCompletionModality
+from openai.types.chat.chat_completion_prediction_content_param import (
+    ChatCompletionPredictionContentParam,
+)
 from openai.types.embedding import Embedding as OpenAIEmbedding
 from pydantic import BaseModel, Field
 from typing_extensions import Dict, Required, TypedDict, override

litellm/utils.py

@@ -2550,6 +2550,7 @@ def get_optional_params(  # noqa: PLR0915
     max_tokens=None,
     max_completion_tokens=None,
     modalities=None,
+    prediction=None,
     audio=None,
     presence_penalty=None,
     frequency_penalty=None,
@@ -2631,6 +2632,7 @@ def get_optional_params(  # noqa: PLR0915
         "max_tokens": None,
         "max_completion_tokens": None,
         "modalities": None,
+        "prediction": None,
         "audio": None,
         "presence_penalty": None,
         "frequency_penalty": None,

pyproject.toml

@@ -17,7 +17,7 @@ documentation = "https://docs.litellm.ai"
 
 [tool.poetry.dependencies]
 python = ">=3.8.1,<4.0, !=3.9.7"
-openai = ">=1.52.0"
+openai = ">=1.54.0"
 python-dotenv = ">=0.2.0"
 tiktoken = ">=0.7.0"
 importlib-metadata = ">=6.8.0"

requirements.txt

@@ -1,6 +1,6 @@
 # LITELLM PROXY DEPENDENCIES #
 anyio==4.4.0 # openai + http req.
-openai==1.52.0 # openai req. 
+openai==1.54.0  # openai req. 
 fastapi==0.111.0 # server dep
 backoff==2.2.1 # server dep
 pyyaml==6.0.0 # server dep