llama-index-llms-openai 0.5.6__py3-none-any.whl → 0.6.0__py3-none-any.whl

llama_index/llms/openai/base.py

@@ -994,7 +994,9 @@ class OpenAI(FunctionCallingLLM):
     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
+        from openai.resources.chat.completions.completions import (
+            _type_to_response_format,
+        )
 
         llm_kwargs = llm_kwargs or {}
         llm_kwargs["response_format"] = _type_to_response_format(output_cls)

llama_index/llms/openai/responses.py

@@ -26,6 +26,7 @@ from openai.types.responses import (
     ResponseReasoningItem,
     ResponseCodeInterpreterToolCall,
     ResponseImageGenCallPartialImageEvent,
+    ResponseOutputItemDoneEvent,
 )
 from openai.types.responses.response_output_item import ImageGenerationCall, McpCall
 from typing import (
@@ -65,6 +66,7 @@ from llama_index.core.base.llms.types import (
     ContentBlock,
     TextBlock,
     ImageBlock,
+    ThinkingBlock,
 )
 from llama_index.core.bridge.pydantic import (
     Field,
@@ -252,7 +254,7 @@ class OpenAIResponses(FunctionCallingLLM):
     default_headers: Optional[Dict[str, str]] = Field(
         default=None, description="The default headers for API requests."
     )
-    api_key: str = Field(default=None, description="The OpenAI API key.")
+    api_key: Optional[str] = Field(default=None, description="The OpenAI API key.")
     api_base: str = Field(description="The base URL for OpenAI API.")
     api_version: str = Field(description="The API version for OpenAI API.")
     context_window: Optional[int] = Field(
@@ -382,7 +384,7 @@ class OpenAIResponses(FunctionCallingLLM):
         return model_name
 
     def _is_azure_client(self) -> bool:
-        return isinstance(self._get_client(), AzureOpenAI)
+        return isinstance(self._client, AzureOpenAI)
 
     def _get_credential_kwargs(self, is_async: bool = False) -> Dict[str, Any]:
         return {
@@ -484,7 +486,22 @@ class OpenAIResponses(FunctionCallingLLM):
             elif isinstance(item, ResponseComputerToolCall):
                 additional_kwargs["built_in_tool_calls"].append(item)
             elif isinstance(item, ResponseReasoningItem):
-                additional_kwargs["reasoning"] = item
+                content: Optional[str] = None
+                if item.content:
+                    content = "\n".join([i.text for i in item.content])
+                if item.summary:
+                    if content:
+                        content += "\n" + "\n".join([i.text for i in item.summary])
+                    else:
+                        content = "\n".join([i.text for i in item.summary])
+                message.blocks.append(
+                    ThinkingBlock(
+                        content=content,
+                        additional_information=item.model_dump(
+                            exclude={"content", "summary"}
+                        ),
+                    )
+                )
 
         if tool_calls and message:
             message.additional_kwargs["tool_calls"] = tool_calls
@@ -512,6 +529,12 @@ class OpenAIResponses(FunctionCallingLLM):
         chat_response = self._parse_response_output(response.output)
         chat_response.raw = response
         chat_response.additional_kwargs["usage"] = response.usage
+        if hasattr(response.usage.output_tokens_details, "reasoning_tokens"):
+            for block in chat_response.message.blocks:
+                if isinstance(block, ThinkingBlock):
+                    block.num_tokens = (
+                        response.usage.output_tokens_details.reasoning_tokens
+                    )
 
         return chat_response
 
@@ -605,9 +628,27 @@ class OpenAIResponses(FunctionCallingLLM):
         elif isinstance(event, ResponseWebSearchCallCompletedEvent):
             # Web search tool call completed
             built_in_tool_calls.append(event)
-        elif isinstance(event, ResponseReasoningItem):
+        elif isinstance(event, ResponseOutputItemDoneEvent):
             # Reasoning information
-            additional_kwargs["reasoning"] = event
+            if isinstance(event.item, ResponseReasoningItem):
+                content: Optional[str] = None
+                if event.item.content:
+                    content = "\n".join([i.text for i in event.item.content])
+                if event.item.summary:
+                    if content:
+                        content += "\n" + "\n".join(
+                            [i.text for i in event.item.summary]
+                        )
+                    else:
+                        content = "\n".join([i.text for i in event.item.summary])
+                blocks.append(
+                    ThinkingBlock(
+                        content=content,
+                        additional_information=event.item.model_dump(
+                            exclude={"content", "summary"}
+                        ),
+                    )
+                )
         elif isinstance(event, ResponseCompletedEvent):
             # Response is complete
             if hasattr(event, "response") and hasattr(event.response, "usage"):

llama_index/llms/openai/utils.py

@@ -29,6 +29,7 @@ from llama_index.core.base.llms.types import (
     TextBlock,
     AudioBlock,
     DocumentBlock,
+    ThinkingBlock,
 )
 from llama_index.core.bridge.pydantic import BaseModel
 
@@ -204,7 +205,7 @@ JSON_SCHEMA_MODELS = [
 
 def is_json_schema_supported(model: str) -> bool:
     try:
-        from openai.resources.beta.chat import completions
+        from openai.resources.chat.completions import completions
 
         if not hasattr(completions, "_type_to_response_format"):
             return False
@@ -469,7 +470,10 @@ def to_openai_responses_message_dict(
 
     for block in message.blocks:
         if isinstance(block, TextBlock):
-            content.append({"type": "input_text", "text": block.text})
+            if message.role.value == "user":
+                content.append({"type": "input_text", "text": block.text})
+            else:
+                content.append({"type": "output_text", "text": block.text})
             content_txt += block.text
         elif isinstance(block, DocumentBlock):
             if not block.data:
@@ -505,6 +509,10 @@ def to_openai_responses_message_dict(
                         "detail": block.detail or "auto",
                     }
                 )
+        elif isinstance(block, ThinkingBlock):
+            if block.content:
+                content.append({"type": "output_text", "text": block.content})
+                content_txt += block.content
         else:
             msg = f"Unsupported content block type: {type(block).__name__}"
             raise ValueError(msg)
@@ -549,9 +557,6 @@ def to_openai_responses_message_dict(
             for tool_call in message.additional_kwargs["tool_calls"]
         ]
 
-        if "reasoning" in message.additional_kwargs:  # and if it is reasoning model
-            message_dicts = [message.additional_kwargs["reasoning"]] + message_dicts
-
         return message_dicts
 
     # there are some cases (like image generation or MCP tool call) that only support the string input

{llama_index_llms_openai-0.5.6.dist-info → llama_index_llms_openai-0.6.0.dist-info}/METADATA

@@ -1,13 +1,13 @@
 Metadata-Version: 2.4
 Name: llama-index-llms-openai
-Version: 0.5.6
+Version: 0.6.0
 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<0.15,>=0.13.0
-Requires-Dist: openai<2,>=1.81.0
+Requires-Dist: openai<2,>=1.108.1
 Description-Content-Type: text/markdown
 
 # LlamaIndex Llms Integration: Openai

llama_index_llms_openai-0.6.0.dist-info/RECORD

@@ -0,0 +1,9 @@
+llama_index/llms/openai/__init__.py,sha256=8nmgixeXifQ4eVSgtCic54WxXqrrpXQPL4rhACWCSFs,229
+llama_index/llms/openai/base.py,sha256=8ckxJJ8LEnmm8_xMTku62HTwAZi8MyjSK5X69PVpums,41900
+llama_index/llms/openai/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+llama_index/llms/openai/responses.py,sha256=zeJ2fTdgf6qAs3VJ7cOFInHAbh_KYogqcQrMzMX71E8,38024
+llama_index/llms/openai/utils.py,sha256=qRoGGpA1FdW9nWPnPMji_FF1vi--RiSqih8x_F8K1C0,30137
+llama_index_llms_openai-0.6.0.dist-info/METADATA,sha256=Er2gjG7VG2JKmNihpK6YoJXmK62S6GtPAH3OJzoxKP8,3039
+llama_index_llms_openai-0.6.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+llama_index_llms_openai-0.6.0.dist-info/licenses/LICENSE,sha256=JPQLUZD9rKvCTdu192Nk0V5PAwklIg6jANii3UmTyMs,1065
+llama_index_llms_openai-0.6.0.dist-info/RECORD,,

llama_index_llms_openai-0.5.6.dist-info/RECORD

@@ -1,9 +0,0 @@
-llama_index/llms/openai/__init__.py,sha256=8nmgixeXifQ4eVSgtCic54WxXqrrpXQPL4rhACWCSFs,229
-llama_index/llms/openai/base.py,sha256=6EANUt_39Pukk6N6CYYVJx6CpoAG2jJsMau6MvV3JTI,41868
-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=3F6TN_IvYw35eCSXKI23EtWf1tScEWzslFS-tGB58Hw,29938
-llama_index_llms_openai-0.5.6.dist-info/METADATA,sha256=9PMeDISUlcyJ2pI2Bw44iqxE93ijoigek14se5PfUjE,3038
-llama_index_llms_openai-0.5.6.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-llama_index_llms_openai-0.5.6.dist-info/licenses/LICENSE,sha256=JPQLUZD9rKvCTdu192Nk0V5PAwklIg6jANii3UmTyMs,1065
-llama_index_llms_openai-0.5.6.dist-info/RECORD,,