llama-index-llms-openai 0.4.3__py3-none-any.whl → 0.4.5__py3-none-any.whl

llama_index/llms/openai/base.py

@@ -77,6 +77,7 @@ from llama_index.llms.openai.utils import (
     resolve_tool_choice,
     to_openai_message_dicts,
     update_tool_calls,
+    is_json_schema_supported,
 )
 from openai import AsyncOpenAI, AzureOpenAI, AsyncAzureOpenAI
 from openai import OpenAI as SyncOpenAI
@@ -1000,6 +1001,23 @@ class OpenAI(FunctionCallingLLM):
 
         return tool_selections
 
+    def _prepare_schema(
+        self, llm_kwargs: Optional[Dict[str, Any]], output_cls: Type[Model]
+    ) -> Dict[str, Any]:
+        from openai.resources.beta.chat.completions import _type_to_response_format
+
+        llm_kwargs = llm_kwargs or {}
+        llm_kwargs["response_format"] = _type_to_response_format(output_cls)
+        if "tool_choice" in llm_kwargs:
+            del llm_kwargs["tool_choice"]
+        return llm_kwargs
+
+    def _should_use_structure_outputs(self):
+        return (
+            self.pydantic_program_mode == PydanticProgramMode.DEFAULT
+            and is_json_schema_supported(self.model)
+        )
+
     @dispatcher.span
     def structured_predict(
         self,
@@ -1011,11 +1029,17 @@ class OpenAI(FunctionCallingLLM):
         """Structured predict."""
         llm_kwargs = llm_kwargs or {}
 
+        if self._should_use_structure_outputs():
+            messages = self._extend_messages(prompt.format_messages(**prompt_args))
+            llm_kwargs = self._prepare_schema(llm_kwargs, output_cls)
+            response = self.chat(messages, **llm_kwargs)
+            return output_cls.model_validate_json(str(response.message.content))
+
+        # when uses function calling to extract structured outputs
+        # here we force tool_choice to be required
         llm_kwargs["tool_choice"] = (
             "required" if "tool_choice" not in llm_kwargs else llm_kwargs["tool_choice"]
         )
-        # by default structured prediction uses function calling to extract structured outputs
-        # here we force tool_choice to be required
         return super().structured_predict(
             output_cls, prompt, llm_kwargs=llm_kwargs, **prompt_args
         )
@@ -1031,15 +1055,91 @@ class OpenAI(FunctionCallingLLM):
         """Structured predict."""
         llm_kwargs = llm_kwargs or {}
 
+        if self._should_use_structure_outputs():
+            messages = self._extend_messages(prompt.format_messages(**prompt_args))
+            llm_kwargs = self._prepare_schema(llm_kwargs, output_cls)
+            response = await self.achat(messages, **llm_kwargs)
+            return output_cls.model_validate_json(str(response.message.content))
+
+        # when uses function calling to extract structured outputs
+        # here we force tool_choice to be required
         llm_kwargs["tool_choice"] = (
             "required" if "tool_choice" not in llm_kwargs else llm_kwargs["tool_choice"]
         )
-        # by default structured prediction uses function calling to extract structured outputs
-        # here we force tool_choice to be required
         return await super().astructured_predict(
             output_cls, prompt, llm_kwargs=llm_kwargs, **prompt_args
         )
 
+    def _structured_stream_call(
+        self,
+        output_cls: Type[Model],
+        prompt: PromptTemplate,
+        llm_kwargs: Optional[Dict[str, Any]] = None,
+        **prompt_args: Any,
+    ) -> Generator[
+        Union[Model, List[Model], "FlexibleModel", List["FlexibleModel"]], None, None
+    ]:
+        if self._should_use_structure_outputs():
+            from llama_index.core.program.streaming_utils import (
+                process_streaming_content_incremental,
+            )
+
+            messages = self._extend_messages(prompt.format_messages(**prompt_args))
+            llm_kwargs = self._prepare_schema(llm_kwargs, output_cls)
+            curr = None
+            for response in self.stream_chat(messages, **llm_kwargs):
+                curr = process_streaming_content_incremental(response, output_cls, curr)
+                yield curr
+        else:
+            llm_kwargs["tool_choice"] = (
+                "required"
+                if "tool_choice" not in llm_kwargs
+                else llm_kwargs["tool_choice"]
+            )
+            yield from super()._structured_stream_call(
+                output_cls, prompt, llm_kwargs, **prompt_args
+            )
+
+    async def _structured_astream_call(
+        self,
+        output_cls: Type[Model],
+        prompt: PromptTemplate,
+        llm_kwargs: Optional[Dict[str, Any]] = None,
+        **prompt_args: Any,
+    ) -> AsyncGenerator[
+        Union[Model, List[Model], "FlexibleModel", List["FlexibleModel"]], None
+    ]:
+        if self._should_use_structure_outputs():
+
+            async def gen(
+                llm_kwargs=llm_kwargs,
+            ) -> AsyncGenerator[
+                Union[Model, List[Model], FlexibleModel, List[FlexibleModel]], None
+            ]:
+                from llama_index.core.program.streaming_utils import (
+                    process_streaming_content_incremental,
+                )
+
+                messages = self._extend_messages(prompt.format_messages(**prompt_args))
+                llm_kwargs = self._prepare_schema(llm_kwargs, output_cls)
+                curr = None
+                async for response in await self.astream_chat(messages, **llm_kwargs):
+                    curr = process_streaming_content_incremental(
+                        response, output_cls, curr
+                    )
+                    yield curr
+
+            return gen()
+        else:
+            llm_kwargs["tool_choice"] = (
+                "required"
+                if "tool_choice" not in llm_kwargs
+                else llm_kwargs["tool_choice"]
+            )
+            return await super()._structured_astream_call(
+                output_cls, prompt, llm_kwargs, **prompt_args
+            )
+
     @dispatcher.span
     def stream_structured_predict(
         self,
@@ -1051,11 +1151,6 @@ class OpenAI(FunctionCallingLLM):
         """Stream structured predict."""
         llm_kwargs = llm_kwargs or {}
 
-        llm_kwargs["tool_choice"] = (
-            "required" if "tool_choice" not in llm_kwargs else llm_kwargs["tool_choice"]
-        )
-        # by default structured prediction uses function calling to extract structured outputs
-        # here we force tool_choice to be required
         return super().stream_structured_predict(
             output_cls, prompt, llm_kwargs=llm_kwargs, **prompt_args
         )
@@ -1070,12 +1165,6 @@ class OpenAI(FunctionCallingLLM):
     ) -> AsyncGenerator[Union[Model, FlexibleModel], None]:
         """Stream structured predict."""
         llm_kwargs = llm_kwargs or {}
-
-        llm_kwargs["tool_choice"] = (
-            "required" if "tool_choice" not in llm_kwargs else llm_kwargs["tool_choice"]
-        )
-        # by default structured prediction uses function calling to extract structured outputs
-        # here we force tool_choice to be required
         return await super().astream_structured_predict(
             output_cls, prompt, llm_kwargs=llm_kwargs, **prompt_args
         )

llama_index/llms/openai/utils.py

@@ -49,6 +49,8 @@ O1_MODELS: Dict[str, int] = {
     "o3-mini-2025-01-31": 200000,
     "o3": 200000,
     "o3-2025-04-16": 200000,
+    "o3-pro": 200000,
+    "o3-pro-2025-06-10": 200000,
     "o4-mini": 200000,
     "o4-mini-2025-04-16": 200000,
 }
@@ -180,6 +182,32 @@ DISCONTINUED_MODELS = {
     "code-cushman-001": 2048,
 }
 
+JSON_SCHEMA_MODELS = [
+    "o4-mini",
+    "o1",
+    "o1-pro",
+    "o3",
+    "o3-mini",
+    "gpt-4.1",
+    "gpt-4o",
+    "gpt-4.1",
+]
+
+
+def is_json_schema_supported(model: str) -> bool:
+    try:
+        from openai.resources.beta.chat import completions
+
+        if not hasattr(completions, "_type_to_response_format"):
+            return False
+
+        return not model.startswith("o1-mini") and any(
+            model.startswith(m) for m in JSON_SCHEMA_MODELS
+        )
+    except ImportError:
+        return False
+
+
 MISSING_API_KEY_ERROR_MESSAGE = """No API key found for OpenAI.
 Please set either the OPENAI_API_KEY environment variable or \
 openai.api_key prior to initialization.

{llama_index_llms_openai-0.4.3.dist-info → llama_index_llms_openai-0.4.5.dist-info}/METADATA

@@ -1,12 +1,12 @@
 Metadata-Version: 2.4
 Name: llama-index-llms-openai
-Version: 0.4.3
+Version: 0.4.5
 Summary: llama-index llms openai integration
 Author: llama-index
 License-Expression: MIT
 License-File: LICENSE
 Requires-Python: <4.0,>=3.9
-Requires-Dist: llama-index-core<0.13,>=0.12.37
+Requires-Dist: llama-index-core<0.13,>=0.12.41
 Requires-Dist: openai<2,>=1.81.0
 Description-Content-Type: text/markdown
 

llama_index_llms_openai-0.4.5.dist-info/RECORD

@@ -0,0 +1,9 @@
+llama_index/llms/openai/__init__.py,sha256=8nmgixeXifQ4eVSgtCic54WxXqrrpXQPL4rhACWCSFs,229
+llama_index/llms/openai/base.py,sha256=DhHPvef7ojecrkk_d7FwA-1Esv2iHymOEkBG0PaE4jg,42282
+llama_index/llms/openai/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+llama_index/llms/openai/responses.py,sha256=Tvlwhst3W_t7EQQCxtj8k-uW6iHmjQPMJ4dh23f-oWg,36182
+llama_index/llms/openai/utils.py,sha256=cYfqKImb_ZwBdT6Jn7gqF84iY014pAvOkbgJt4LOfUc,28644
+llama_index_llms_openai-0.4.5.dist-info/METADATA,sha256=J9M-2u-13ZR_hhz6BIS9Bs5lJWlm7TblIGvxOwNyCY8,3039
+llama_index_llms_openai-0.4.5.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+llama_index_llms_openai-0.4.5.dist-info/licenses/LICENSE,sha256=JPQLUZD9rKvCTdu192Nk0V5PAwklIg6jANii3UmTyMs,1065
+llama_index_llms_openai-0.4.5.dist-info/RECORD,,

llama_index_llms_openai-0.4.3.dist-info/RECORD

@@ -1,9 +0,0 @@
-llama_index/llms/openai/__init__.py,sha256=8nmgixeXifQ4eVSgtCic54WxXqrrpXQPL4rhACWCSFs,229
-llama_index/llms/openai/base.py,sha256=LTed9nxKpEnfha_FNDBrzNP3B1CTDFwaxGyAFUXXV5o,38856
-llama_index/llms/openai/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-llama_index/llms/openai/responses.py,sha256=Tvlwhst3W_t7EQQCxtj8k-uW6iHmjQPMJ4dh23f-oWg,36182
-llama_index/llms/openai/utils.py,sha256=0xO-_6FPZRFCbW5JeqJDCMzLx3WN2_vVeAApWG74oA4,28071
-llama_index_llms_openai-0.4.3.dist-info/METADATA,sha256=L2d4szZKIinQcdHiojQBj52NK7c63eOuOgx9Id4-Hpg,3039
-llama_index_llms_openai-0.4.3.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-llama_index_llms_openai-0.4.3.dist-info/licenses/LICENSE,sha256=JPQLUZD9rKvCTdu192Nk0V5PAwklIg6jANii3UmTyMs,1065
-llama_index_llms_openai-0.4.3.dist-info/RECORD,,