OpenAI Structured Output

CHANGELOG.md

@@ -23,6 +23,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   - `LocalRulesetDriver` for loading a `Ruleset` from a local `.json` file.
   - `GriptapeCloudRulesetDriver` for loading a `Ruleset` resource from Griptape Cloud.
 - Parameter `alias` on `GriptapeCloudConversationMemoryDriver` for fetching a Thread by alias.
+- Basic support for OpenAi Structured Output via `OpenAiChatPromptDriver.response_format` parameter. 
 
 ### Changed
 - **BREAKING**: Renamed parameters on several classes to `client`:
@@ -61,6 +62,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **BREAKING**: `CsvExtractionEngine.column_names` is now required.
 - **BREAKING**: Renamed`RuleMixin.all_rulesets` to `RuleMixin.rulesets`.
 - **BREAKING**: Renamed `GriptapeCloudKnowledgeBaseVectorStoreDriver` to `GriptapeCloudVectorStoreDriver`.
+- **BREAKING**: `OpenAiChatPromptDriver.response_format` is now a `dict` instead of a `str`.
 - `MarkdownifyWebScraperDriver.DEFAULT_EXCLUDE_TAGS` now includes media/blob-like HTML tags
 - `StructureRunTask` now inherits from `PromptTask`.
 - Several places where API clients are initialized are now lazy loaded.

MIGRATION.md

@@ -168,6 +168,24 @@ from griptape.drivers.griptape_cloud_vector_store_driver import GriptapeCloudVec
 driver = GriptapeCloudVectorStoreDriver(...)
 ```
 
+### `OpenAiChatPromptDriver.response_format` is now a `dict` instead of a `str`.
+
+`OpenAiChatPromptDriver.response_format` is now structured as the `openai` SDK accepts it.
+
+#### Before
+```python
+driver = OpenAiChatPromptDriver(
+    response_format="json_object"
+)
+```
+
+#### After
+```python
+driver = OpenAiChatPromptDriver(
+    response_format={"type": "json_object"}
+)
+```
+
 ## 0.31.X to 0.32.X
 
 ### Removed `DataframeLoader`

docs/griptape-framework/drivers/src/prompt_drivers_3.py

@@ -1,19 +1,26 @@
 import os
 
+import schema
+
 from griptape.drivers import OpenAiChatPromptDriver
-from griptape.rules import Rule
 from griptape.structures import Agent
 
 agent = Agent(
     prompt_driver=OpenAiChatPromptDriver(
         api_key=os.environ["OPENAI_API_KEY"],
+        model="gpt-4o-2024-08-06",
         temperature=0.1,
-        model="gpt-4o",
-        response_format="json_object",
         seed=42,
+        response_format={
+            "type": "json_schema",
+            "json_schema": {
+                "strict": True,
+                "name": "Output",
+                "schema": schema.Schema({"css_code": str, "relevant_emojies": [str]}).json_schema("Output Schema"),
+            },
+        },
     ),
-    input="You will be provided with a description of a mood, and your task is to generate the CSS code for a color that matches it. Description: {{ args[0] }}",
-    rules=[Rule(value='Write your output in json with a single key called "css_code".')],
+    input="You will be provided with a description of a mood, and your task is to generate the CSS color code for a color that matches it. Description: {{ args[0] }}",
 )
 
 agent.run("Blue sky at dusk.")

griptape/drivers/prompt/openai_chat_prompt_driver.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 
 import json
-from typing import TYPE_CHECKING, Literal, Optional
+from typing import TYPE_CHECKING, Optional
 
 import openai
 from attrs import Factory, define, field
@@ -62,7 +62,7 @@ class OpenAiChatPromptDriver(BasePromptDriver):
         kw_only=True,
     )
     user: str = field(default="", kw_only=True, metadata={"serializable": True})
-    response_format: Optional[Literal["json_object"]] = field(
+    response_format: Optional[dict] = field(
         default=None,
         kw_only=True,
         metadata={"serializable": True},
@@ -145,10 +145,13 @@ def _base_params(self, prompt_stack: PromptStack) -> dict:
             **({"stream_options": {"include_usage": True}} if self.stream else {}),
         }
 
-        if self.response_format == "json_object":
-            params["response_format"] = {"type": "json_object"}
-            # JSON mode still requires a system message instructing the LLM to output JSON.
-            prompt_stack.add_system_message("Provide your response as a valid JSON object.")
+        if self.response_format is not None:
+            if self.response_format == {"type": "json_object"}:
+                params["response_format"] = self.response_format
+                # JSON mode still requires a system message instructing the LLM to output JSON.
+                prompt_stack.add_system_message("Provide your response as a valid JSON object.")
+            else:
+                params["response_format"] = self.response_format
 
         messages = self.__to_openai_messages(prompt_stack.messages)
 

tests/unit/drivers/prompt/test_openai_chat_prompt_driver.py

@@ -1,6 +1,7 @@
 from unittest.mock import Mock
 
 import pytest
+import schema
 
 from griptape.artifacts import ActionArtifact, ImageArtifact, ListArtifact, TextArtifact
 from griptape.common import ActionCallDeltaMessageContent, PromptStack, TextDeltaMessageContent, ToolAction
@@ -343,10 +344,12 @@ def test_try_run(self, mock_chat_completion_create, prompt_stack, messages, use_
         assert message.value[1].value.path == "test"
         assert message.value[1].value.input == {"foo": "bar"}
 
-    def test_try_run_response_format(self, mock_chat_completion_create, prompt_stack, messages):
+    def test_try_run_response_format_json_object(self, mock_chat_completion_create, prompt_stack, messages):
         # Given
         driver = OpenAiChatPromptDriver(
-            model=OpenAiTokenizer.DEFAULT_OPENAI_GPT_3_CHAT_MODEL, response_format="json_object", use_native_tools=False
+            model=OpenAiTokenizer.DEFAULT_OPENAI_GPT_3_CHAT_MODEL,
+            response_format={"type": "json_object"},
+            use_native_tools=False,
         )
 
         # When
@@ -365,6 +368,51 @@ def test_try_run_response_format(self, mock_chat_completion_create, prompt_stack
         assert message.usage.input_tokens == 5
         assert message.usage.output_tokens == 10
 
+    def test_try_run_response_format_json_schema(self, mock_chat_completion_create, prompt_stack, messages):
+        # Given
+        driver = OpenAiChatPromptDriver(
+            model=OpenAiTokenizer.DEFAULT_OPENAI_GPT_3_CHAT_MODEL,
+            response_format={
+                "type": "json_schema",
+                "json_schema": {
+                    "strict": True,
+                    "name": "OutputSchema",
+                    "schema": schema.Schema({"test": str}).json_schema("Output Schema"),
+                },
+            },
+            use_native_tools=False,
+        )
+
+        # When
+        message = driver.try_run(prompt_stack)
+
+        # Then
+        mock_chat_completion_create.assert_called_once_with(
+            model=driver.model,
+            temperature=driver.temperature,
+            user=driver.user,
+            messages=[*messages],
+            seed=driver.seed,
+            response_format={
+                "json_schema": {
+                    "schema": {
+                        "$id": "Output Schema",
+                        "$schema": "http://json-schema.org/draft-07/schema#",
+                        "additionalProperties": False,
+                        "properties": {"test": {"type": "string"}},
+                        "required": ["test"],
+                        "type": "object",
+                    },
+                    "name": "OutputSchema",
+                    "strict": True,
+                },
+                "type": "json_schema",
+            },
+        )
+        assert message.value[0].value == "model-output"
+        assert message.usage.input_tokens == 5
+        assert message.usage.output_tokens == 10
+
     @pytest.mark.parametrize("use_native_tools", [True, False])
     def test_try_stream_run(self, mock_chat_completion_stream_create, prompt_stack, messages, use_native_tools):
         # Given

tests/utils/structure_tester.py

@@ -235,7 +235,7 @@ def verify_structure_output(self, structure) -> dict:
             model="gpt-4o",
             azure_deployment=os.environ["AZURE_OPENAI_4_DEPLOYMENT_ID"],
             azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT_1"],
-            response_format="json_object",
+            response_format={"type": "json_object"},
         )
         output_schema = Schema(
             {