mistralai 1.9.11__py3-none-any.whl → 1.10.1__py3-none-any.whl

mistralai_azure/basesdk.py

@@ -15,9 +15,19 @@ from urllib.parse import parse_qs, urlparse
 
 class BaseSDK:
     sdk_configuration: SDKConfiguration
+    parent_ref: Optional[object] = None
+    """
+    Reference to the root SDK instance, if any. This will prevent it from
+    being garbage collected while there are active streams.
+    """
 
-    def __init__(self, sdk_config: SDKConfiguration) -> None:
+    def __init__(
+        self,
+        sdk_config: SDKConfiguration,
+        parent_ref: Optional[object] = None,
+    ) -> None:
         self.sdk_configuration = sdk_config
+        self.parent_ref = parent_ref
 
     def _get_url(self, base_url, url_variables):
         sdk_url, sdk_variables = self.sdk_configuration.get_server_details()
@@ -244,7 +254,7 @@ class BaseSDK:
 
             if http_res is None:
                 logger.debug("Raising no response SDK error")
-                raise models.SDKError("No response received")
+                raise models.NoResponseError("No response received")
 
             logger.debug(
                 "Response:\nStatus Code: %s\nURL: %s\nHeaders: %s\nBody: %s",
@@ -265,7 +275,7 @@ class BaseSDK:
                     http_res = result
                 else:
                     logger.debug("Raising unexpected SDK error")
-                    raise models.SDKError("Unexpected error occurred")
+                    raise models.SDKError("Unexpected error occurred", http_res)
 
             return http_res
 
@@ -316,7 +326,7 @@ class BaseSDK:
 
             if http_res is None:
                 logger.debug("Raising no response SDK error")
-                raise models.SDKError("No response received")
+                raise models.NoResponseError("No response received")
 
             logger.debug(
                 "Response:\nStatus Code: %s\nURL: %s\nHeaders: %s\nBody: %s",
@@ -337,7 +347,7 @@ class BaseSDK:
                     http_res = result
                 else:
                     logger.debug("Raising unexpected SDK error")
-                    raise models.SDKError("Unexpected error occurred")
+                    raise models.SDKError("Unexpected error occurred", http_res)
 
             return http_res
 

mistralai_azure/chat.py

@@ -5,6 +5,7 @@ from mistralai_azure import models, utils
 from mistralai_azure._hooks import HookContext
 from mistralai_azure.types import OptionalNullable, UNSET
 from mistralai_azure.utils import eventstreaming
+from mistralai_azure.utils.unmarshal_json_response import unmarshal_json_response
 from typing import Any, List, Mapping, Optional, Union
 
 
@@ -60,14 +61,14 @@ class Chat(BaseSDK):
         :param stream:
         :param stop: Stop generation if this token is detected. Or if one of these tokens is detected when providing an array
         :param random_seed: The seed to use for random sampling. If set, different calls will generate deterministic results.
-        :param response_format:
-        :param tools:
-        :param tool_choice:
-        :param presence_penalty: presence_penalty determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative.
-        :param frequency_penalty: frequency_penalty penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition.
+        :param response_format: Specify the format that the model must output. By default it will use `{ \"type\": \"text\" }`. Setting to `{ \"type\": \"json_object\" }` enables JSON mode, which guarantees the message the model generates is in JSON. When using JSON mode you MUST also instruct the model to produce JSON yourself with a system or a user message. Setting to `{ \"type\": \"json_schema\" }` enables JSON schema mode, which guarantees the message the model generates is in JSON and follows the schema you provide.
+        :param tools: A list of tools the model may call. Use this to provide a list of functions the model may generate JSON inputs for.
+        :param tool_choice: Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool.
+        :param presence_penalty: The `presence_penalty` determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative.
+        :param frequency_penalty: The `frequency_penalty` penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition.
         :param n: Number of completions to return for each request, input tokens are only billed once.
-        :param prediction:
-        :param parallel_tool_calls:
+        :param prediction: Enable users to specify an expected completion, optimizing response times by leveraging known or predictable content.
+        :param parallel_tool_calls: Whether to enable parallel function calling during tool use, when enabled the model can call multiple tools in parallel.
         :param prompt_mode: Allows toggling between the reasoning mode and no system prompt. When set to `reasoning` the system prompt for reasoning models will be used.
         :param safe_prompt: Whether to inject a safety prompt before all conversations.
         :param retries: Override the default retry configuration for this method
@@ -159,32 +160,23 @@ class Chat(BaseSDK):
                 http_res,
                 lambda raw: utils.unmarshal_json(raw, models.CompletionEvent),
                 sentinel="[DONE]",
+                client_ref=self,
             )
         if utils.match_response(http_res, "422", "application/json"):
             http_res_text = utils.stream_to_text(http_res)
-            response_data = utils.unmarshal_json(
-                http_res_text, models.HTTPValidationErrorData
+            response_data = unmarshal_json_response(
+                models.HTTPValidationErrorData, http_res, http_res_text
             )
-            raise models.HTTPValidationError(data=response_data)
+            raise models.HTTPValidationError(response_data, http_res, http_res_text)
         if utils.match_response(http_res, "4XX", "*"):
             http_res_text = utils.stream_to_text(http_res)
-            raise models.SDKError(
-                "API error occurred", http_res.status_code, http_res_text, http_res
-            )
+            raise models.SDKError("API error occurred", http_res, http_res_text)
         if utils.match_response(http_res, "5XX", "*"):
             http_res_text = utils.stream_to_text(http_res)
-            raise models.SDKError(
-                "API error occurred", http_res.status_code, http_res_text, http_res
-            )
+            raise models.SDKError("API error occurred", http_res, http_res_text)
 
-        content_type = http_res.headers.get("Content-Type")
         http_res_text = utils.stream_to_text(http_res)
-        raise models.SDKError(
-            f"Unexpected response received (code: {http_res.status_code}, type: {content_type})",
-            http_res.status_code,
-            http_res_text,
-            http_res,
-        )
+        raise models.SDKError("Unexpected response received", http_res, http_res_text)
 
     async def stream_async(
         self,
@@ -235,14 +227,14 @@ class Chat(BaseSDK):
         :param stream:
         :param stop: Stop generation if this token is detected. Or if one of these tokens is detected when providing an array
         :param random_seed: The seed to use for random sampling. If set, different calls will generate deterministic results.
-        :param response_format:
-        :param tools:
-        :param tool_choice:
-        :param presence_penalty: presence_penalty determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative.
-        :param frequency_penalty: frequency_penalty penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition.
+        :param response_format: Specify the format that the model must output. By default it will use `{ \"type\": \"text\" }`. Setting to `{ \"type\": \"json_object\" }` enables JSON mode, which guarantees the message the model generates is in JSON. When using JSON mode you MUST also instruct the model to produce JSON yourself with a system or a user message. Setting to `{ \"type\": \"json_schema\" }` enables JSON schema mode, which guarantees the message the model generates is in JSON and follows the schema you provide.
+        :param tools: A list of tools the model may call. Use this to provide a list of functions the model may generate JSON inputs for.
+        :param tool_choice: Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool.
+        :param presence_penalty: The `presence_penalty` determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative.
+        :param frequency_penalty: The `frequency_penalty` penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition.
         :param n: Number of completions to return for each request, input tokens are only billed once.
-        :param prediction:
-        :param parallel_tool_calls:
+        :param prediction: Enable users to specify an expected completion, optimizing response times by leveraging known or predictable content.
+        :param parallel_tool_calls: Whether to enable parallel function calling during tool use, when enabled the model can call multiple tools in parallel.
         :param prompt_mode: Allows toggling between the reasoning mode and no system prompt. When set to `reasoning` the system prompt for reasoning models will be used.
         :param safe_prompt: Whether to inject a safety prompt before all conversations.
         :param retries: Override the default retry configuration for this method
@@ -334,32 +326,23 @@ class Chat(BaseSDK):
                 http_res,
                 lambda raw: utils.unmarshal_json(raw, models.CompletionEvent),
                 sentinel="[DONE]",
+                client_ref=self,
             )
         if utils.match_response(http_res, "422", "application/json"):
             http_res_text = await utils.stream_to_text_async(http_res)
-            response_data = utils.unmarshal_json(
-                http_res_text, models.HTTPValidationErrorData
+            response_data = unmarshal_json_response(
+                models.HTTPValidationErrorData, http_res, http_res_text
             )
-            raise models.HTTPValidationError(data=response_data)
+            raise models.HTTPValidationError(response_data, http_res, http_res_text)
         if utils.match_response(http_res, "4XX", "*"):
             http_res_text = await utils.stream_to_text_async(http_res)
-            raise models.SDKError(
-                "API error occurred", http_res.status_code, http_res_text, http_res
-            )
+            raise models.SDKError("API error occurred", http_res, http_res_text)
         if utils.match_response(http_res, "5XX", "*"):
             http_res_text = await utils.stream_to_text_async(http_res)
-            raise models.SDKError(
-                "API error occurred", http_res.status_code, http_res_text, http_res
-            )
+            raise models.SDKError("API error occurred", http_res, http_res_text)
 
-        content_type = http_res.headers.get("Content-Type")
         http_res_text = await utils.stream_to_text_async(http_res)
-        raise models.SDKError(
-            f"Unexpected response received (code: {http_res.status_code}, type: {content_type})",
-            http_res.status_code,
-            http_res_text,
-            http_res,
-        )
+        raise models.SDKError("Unexpected response received", http_res, http_res_text)
 
     def complete(
         self,
@@ -416,14 +399,14 @@ class Chat(BaseSDK):
         :param stream: Whether to stream back partial progress. If set, tokens will be sent as data-only server-side events as they become available, with the stream terminated by a data: [DONE] message. Otherwise, the server will hold the request open until the timeout or until completion, with the response containing the full result as JSON.
         :param stop: Stop generation if this token is detected. Or if one of these tokens is detected when providing an array
         :param random_seed: The seed to use for random sampling. If set, different calls will generate deterministic results.
-        :param response_format:
-        :param tools:
-        :param tool_choice:
-        :param presence_penalty: presence_penalty determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative.
-        :param frequency_penalty: frequency_penalty penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition.
+        :param response_format: Specify the format that the model must output. By default it will use `{ \"type\": \"text\" }`. Setting to `{ \"type\": \"json_object\" }` enables JSON mode, which guarantees the message the model generates is in JSON. When using JSON mode you MUST also instruct the model to produce JSON yourself with a system or a user message. Setting to `{ \"type\": \"json_schema\" }` enables JSON schema mode, which guarantees the message the model generates is in JSON and follows the schema you provide.
+        :param tools: A list of tools the model may call. Use this to provide a list of functions the model may generate JSON inputs for.
+        :param tool_choice: Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool.
+        :param presence_penalty: The `presence_penalty` determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative.
+        :param frequency_penalty: The `frequency_penalty` penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition.
         :param n: Number of completions to return for each request, input tokens are only billed once.
-        :param prediction:
-        :param parallel_tool_calls:
+        :param prediction: Enable users to specify an expected completion, optimizing response times by leveraging known or predictable content.
+        :param parallel_tool_calls: Whether to enable parallel function calling during tool use, when enabled the model can call multiple tools in parallel.
         :param prompt_mode: Allows toggling between the reasoning mode and no system prompt. When set to `reasoning` the system prompt for reasoning models will be used.
         :param safe_prompt: Whether to inject a safety prompt before all conversations.
         :param retries: Override the default retry configuration for this method
@@ -512,33 +495,22 @@ class Chat(BaseSDK):
 
         response_data: Any = None
         if utils.match_response(http_res, "200", "application/json"):
-            return utils.unmarshal_json(
-                http_res.text, Optional[models.ChatCompletionResponse]
+            return unmarshal_json_response(
+                Optional[models.ChatCompletionResponse], http_res
             )
         if utils.match_response(http_res, "422", "application/json"):
-            response_data = utils.unmarshal_json(
-                http_res.text, models.HTTPValidationErrorData
+            response_data = unmarshal_json_response(
+                models.HTTPValidationErrorData, http_res
             )
-            raise models.HTTPValidationError(data=response_data)
+            raise models.HTTPValidationError(response_data, http_res)
         if utils.match_response(http_res, "4XX", "*"):
             http_res_text = utils.stream_to_text(http_res)
-            raise models.SDKError(
-                "API error occurred", http_res.status_code, http_res_text, http_res
-            )
+            raise models.SDKError("API error occurred", http_res, http_res_text)
         if utils.match_response(http_res, "5XX", "*"):
             http_res_text = utils.stream_to_text(http_res)
-            raise models.SDKError(
-                "API error occurred", http_res.status_code, http_res_text, http_res
-            )
+            raise models.SDKError("API error occurred", http_res, http_res_text)
 
-        content_type = http_res.headers.get("Content-Type")
-        http_res_text = utils.stream_to_text(http_res)
-        raise models.SDKError(
-            f"Unexpected response received (code: {http_res.status_code}, type: {content_type})",
-            http_res.status_code,
-            http_res_text,
-            http_res,
-        )
+        raise models.SDKError("Unexpected response received", http_res)
 
     async def complete_async(
         self,
@@ -595,14 +567,14 @@ class Chat(BaseSDK):
         :param stream: Whether to stream back partial progress. If set, tokens will be sent as data-only server-side events as they become available, with the stream terminated by a data: [DONE] message. Otherwise, the server will hold the request open until the timeout or until completion, with the response containing the full result as JSON.
         :param stop: Stop generation if this token is detected. Or if one of these tokens is detected when providing an array
         :param random_seed: The seed to use for random sampling. If set, different calls will generate deterministic results.
-        :param response_format:
-        :param tools:
-        :param tool_choice:
-        :param presence_penalty: presence_penalty determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative.
-        :param frequency_penalty: frequency_penalty penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition.
+        :param response_format: Specify the format that the model must output. By default it will use `{ \"type\": \"text\" }`. Setting to `{ \"type\": \"json_object\" }` enables JSON mode, which guarantees the message the model generates is in JSON. When using JSON mode you MUST also instruct the model to produce JSON yourself with a system or a user message. Setting to `{ \"type\": \"json_schema\" }` enables JSON schema mode, which guarantees the message the model generates is in JSON and follows the schema you provide.
+        :param tools: A list of tools the model may call. Use this to provide a list of functions the model may generate JSON inputs for.
+        :param tool_choice: Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool.
+        :param presence_penalty: The `presence_penalty` determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative.
+        :param frequency_penalty: The `frequency_penalty` penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition.
         :param n: Number of completions to return for each request, input tokens are only billed once.
-        :param prediction:
-        :param parallel_tool_calls:
+        :param prediction: Enable users to specify an expected completion, optimizing response times by leveraging known or predictable content.
+        :param parallel_tool_calls: Whether to enable parallel function calling during tool use, when enabled the model can call multiple tools in parallel.
         :param prompt_mode: Allows toggling between the reasoning mode and no system prompt. When set to `reasoning` the system prompt for reasoning models will be used.
         :param safe_prompt: Whether to inject a safety prompt before all conversations.
         :param retries: Override the default retry configuration for this method
@@ -691,30 +663,19 @@ class Chat(BaseSDK):
 
         response_data: Any = None
         if utils.match_response(http_res, "200", "application/json"):
-            return utils.unmarshal_json(
-                http_res.text, Optional[models.ChatCompletionResponse]
+            return unmarshal_json_response(
+                Optional[models.ChatCompletionResponse], http_res
             )
         if utils.match_response(http_res, "422", "application/json"):
-            response_data = utils.unmarshal_json(
-                http_res.text, models.HTTPValidationErrorData
+            response_data = unmarshal_json_response(
+                models.HTTPValidationErrorData, http_res
             )
-            raise models.HTTPValidationError(data=response_data)
+            raise models.HTTPValidationError(response_data, http_res)
         if utils.match_response(http_res, "4XX", "*"):
             http_res_text = await utils.stream_to_text_async(http_res)
-            raise models.SDKError(
-                "API error occurred", http_res.status_code, http_res_text, http_res
-            )
+            raise models.SDKError("API error occurred", http_res, http_res_text)
         if utils.match_response(http_res, "5XX", "*"):
             http_res_text = await utils.stream_to_text_async(http_res)
-            raise models.SDKError(
-                "API error occurred", http_res.status_code, http_res_text, http_res
-            )
+            raise models.SDKError("API error occurred", http_res, http_res_text)
 
-        content_type = http_res.headers.get("Content-Type")
-        http_res_text = await utils.stream_to_text_async(http_res)
-        raise models.SDKError(
-            f"Unexpected response received (code: {http_res.status_code}, type: {content_type})",
-            http_res.status_code,
-            http_res_text,
-            http_res,
-        )
+        raise models.SDKError("Unexpected response received", http_res)

mistralai_azure/models/__init__.py

@@ -1,7 +1,10 @@
 """Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT."""
 
+from .mistralazureerror import MistralAzureError
 from typing import TYPE_CHECKING
 from importlib import import_module
+import builtins
+import sys
 
 if TYPE_CHECKING:
     from .assistantmessage import (
@@ -79,6 +82,7 @@ if TYPE_CHECKING:
     )
     from .jsonschema import JSONSchema, JSONSchemaTypedDict
     from .mistralpromptmode import MistralPromptMode
+    from .no_response_error import NoResponseError
     from .ocrimageobject import OCRImageObject, OCRImageObjectTypedDict
     from .ocrpagedimensions import OCRPageDimensions, OCRPageDimensionsTypedDict
     from .ocrpageobject import OCRPageObject, OCRPageObjectTypedDict
@@ -93,6 +97,7 @@ if TYPE_CHECKING:
     )
     from .responseformat import ResponseFormat, ResponseFormatTypedDict
     from .responseformats import ResponseFormats
+    from .responsevalidationerror import ResponseValidationError
     from .sdkerror import SDKError
     from .security import Security, SecurityTypedDict
     from .systemmessage import (
@@ -102,7 +107,18 @@ if TYPE_CHECKING:
         SystemMessageContentTypedDict,
         SystemMessageTypedDict,
     )
+    from .systemmessagecontentchunks import (
+        SystemMessageContentChunks,
+        SystemMessageContentChunksTypedDict,
+    )
     from .textchunk import TextChunk, TextChunkTypedDict, Type
+    from .thinkchunk import (
+        ThinkChunk,
+        ThinkChunkType,
+        ThinkChunkTypedDict,
+        Thinking,
+        ThinkingTypedDict,
+    )
     from .tool import Tool, ToolTypedDict
     from .toolcall import ToolCall, ToolCallTypedDict
     from .toolchoice import ToolChoice, ToolChoiceTypedDict
@@ -196,7 +212,9 @@ __all__ = [
     "LocTypedDict",
     "Messages",
     "MessagesTypedDict",
+    "MistralAzureError",
     "MistralPromptMode",
+    "NoResponseError",
     "OCRImageObject",
     "OCRImageObjectTypedDict",
     "OCRPageDimensions",
@@ -217,6 +235,7 @@ __all__ = [
     "ResponseFormat",
     "ResponseFormatTypedDict",
     "ResponseFormats",
+    "ResponseValidationError",
     "Role",
     "SDKError",
     "Security",
@@ -225,10 +244,17 @@ __all__ = [
     "StopTypedDict",
     "SystemMessage",
     "SystemMessageContent",
+    "SystemMessageContentChunks",
+    "SystemMessageContentChunksTypedDict",
     "SystemMessageContentTypedDict",
     "SystemMessageTypedDict",
     "TextChunk",
     "TextChunkTypedDict",
+    "ThinkChunk",
+    "ThinkChunkType",
+    "ThinkChunkTypedDict",
+    "Thinking",
+    "ThinkingTypedDict",
     "Tool",
     "ToolCall",
     "ToolCallTypedDict",
@@ -319,6 +345,7 @@ _dynamic_imports: dict[str, str] = {
     "JSONSchema": ".jsonschema",
     "JSONSchemaTypedDict": ".jsonschema",
     "MistralPromptMode": ".mistralpromptmode",
+    "NoResponseError": ".no_response_error",
     "OCRImageObject": ".ocrimageobject",
     "OCRImageObjectTypedDict": ".ocrimageobject",
     "OCRPageDimensions": ".ocrpagedimensions",
@@ -341,6 +368,7 @@ _dynamic_imports: dict[str, str] = {
     "ResponseFormat": ".responseformat",
     "ResponseFormatTypedDict": ".responseformat",
     "ResponseFormats": ".responseformats",
+    "ResponseValidationError": ".responsevalidationerror",
     "SDKError": ".sdkerror",
     "Security": ".security",
     "SecurityTypedDict": ".security",
@@ -349,9 +377,16 @@ _dynamic_imports: dict[str, str] = {
     "SystemMessageContent": ".systemmessage",
     "SystemMessageContentTypedDict": ".systemmessage",
     "SystemMessageTypedDict": ".systemmessage",
+    "SystemMessageContentChunks": ".systemmessagecontentchunks",
+    "SystemMessageContentChunksTypedDict": ".systemmessagecontentchunks",
     "TextChunk": ".textchunk",
     "TextChunkTypedDict": ".textchunk",
     "Type": ".textchunk",
+    "ThinkChunk": ".thinkchunk",
+    "ThinkChunkType": ".thinkchunk",
+    "ThinkChunkTypedDict": ".thinkchunk",
+    "Thinking": ".thinkchunk",
+    "ThinkingTypedDict": ".thinkchunk",
     "Tool": ".tool",
     "ToolTypedDict": ".tool",
     "ToolCall": ".toolcall",
@@ -379,6 +414,18 @@ _dynamic_imports: dict[str, str] = {
 }
 
 
+def dynamic_import(modname, retries=3):
+    for attempt in range(retries):
+        try:
+            return import_module(modname, __package__)
+        except KeyError:
+            # Clear any half-initialized module and retry
+            sys.modules.pop(modname, None)
+            if attempt == retries - 1:
+                break
+    raise KeyError(f"Failed to import module '{modname}' after {retries} attempts")
+
+
 def __getattr__(attr_name: str) -> object:
     module_name = _dynamic_imports.get(attr_name)
     if module_name is None:
@@ -387,7 +434,7 @@ def __getattr__(attr_name: str) -> object:
         )
 
     try:
-        module = import_module(module_name, __package__)
+        module = dynamic_import(module_name)
         result = getattr(module, attr_name)
         return result
     except ImportError as e:
@@ -401,5 +448,5 @@ def __getattr__(attr_name: str) -> object:
 
 
 def __dir__():
-    lazy_attrs = list(_dynamic_imports.keys())
-    return sorted(lazy_attrs)
+    lazy_attrs = builtins.list(_dynamic_imports.keys())
+    return builtins.sorted(lazy_attrs)

mistralai_azure/models/chatcompletionrequest.py

@@ -63,11 +63,13 @@ ChatCompletionRequestToolChoiceTypedDict = TypeAliasType(
     "ChatCompletionRequestToolChoiceTypedDict",
     Union[ToolChoiceTypedDict, ToolChoiceEnum],
 )
+r"""Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool."""
 
 
 ChatCompletionRequestToolChoice = TypeAliasType(
     "ChatCompletionRequestToolChoice", Union[ToolChoice, ToolChoiceEnum]
 )
+r"""Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool."""
 
 
 class ChatCompletionRequestTypedDict(TypedDict):
@@ -88,16 +90,21 @@ class ChatCompletionRequestTypedDict(TypedDict):
     random_seed: NotRequired[Nullable[int]]
     r"""The seed to use for random sampling. If set, different calls will generate deterministic results."""
     response_format: NotRequired[ResponseFormatTypedDict]
+    r"""Specify the format that the model must output. By default it will use `{ \"type\": \"text\" }`. Setting to `{ \"type\": \"json_object\" }` enables JSON mode, which guarantees the message the model generates is in JSON. When using JSON mode you MUST also instruct the model to produce JSON yourself with a system or a user message. Setting to `{ \"type\": \"json_schema\" }` enables JSON schema mode, which guarantees the message the model generates is in JSON and follows the schema you provide."""
     tools: NotRequired[Nullable[List[ToolTypedDict]]]
+    r"""A list of tools the model may call. Use this to provide a list of functions the model may generate JSON inputs for."""
     tool_choice: NotRequired[ChatCompletionRequestToolChoiceTypedDict]
+    r"""Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool."""
     presence_penalty: NotRequired[float]
-    r"""presence_penalty determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative."""
+    r"""The `presence_penalty` determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative."""
     frequency_penalty: NotRequired[float]
-    r"""frequency_penalty penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition."""
+    r"""The `frequency_penalty` penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition."""
     n: NotRequired[Nullable[int]]
     r"""Number of completions to return for each request, input tokens are only billed once."""
     prediction: NotRequired[PredictionTypedDict]
+    r"""Enable users to specify an expected completion, optimizing response times by leveraging known or predictable content."""
     parallel_tool_calls: NotRequired[bool]
+    r"""Whether to enable parallel function calling during tool use, when enabled the model can call multiple tools in parallel."""
     prompt_mode: NotRequired[Nullable[MistralPromptMode]]
     r"""Allows toggling between the reasoning mode and no system prompt. When set to `reasoning` the system prompt for reasoning models will be used."""
     safe_prompt: NotRequired[bool]
@@ -130,23 +137,28 @@ class ChatCompletionRequest(BaseModel):
     r"""The seed to use for random sampling. If set, different calls will generate deterministic results."""
 
     response_format: Optional[ResponseFormat] = None
+    r"""Specify the format that the model must output. By default it will use `{ \"type\": \"text\" }`. Setting to `{ \"type\": \"json_object\" }` enables JSON mode, which guarantees the message the model generates is in JSON. When using JSON mode you MUST also instruct the model to produce JSON yourself with a system or a user message. Setting to `{ \"type\": \"json_schema\" }` enables JSON schema mode, which guarantees the message the model generates is in JSON and follows the schema you provide."""
 
     tools: OptionalNullable[List[Tool]] = UNSET
+    r"""A list of tools the model may call. Use this to provide a list of functions the model may generate JSON inputs for."""
 
     tool_choice: Optional[ChatCompletionRequestToolChoice] = None
+    r"""Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool."""
 
     presence_penalty: Optional[float] = None
-    r"""presence_penalty determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative."""
+    r"""The `presence_penalty` determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative."""
 
     frequency_penalty: Optional[float] = None
-    r"""frequency_penalty penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition."""
+    r"""The `frequency_penalty` penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition."""
 
     n: OptionalNullable[int] = UNSET
     r"""Number of completions to return for each request, input tokens are only billed once."""
 
     prediction: Optional[Prediction] = None
+    r"""Enable users to specify an expected completion, optimizing response times by leveraging known or predictable content."""
 
     parallel_tool_calls: Optional[bool] = None
+    r"""Whether to enable parallel function calling during tool use, when enabled the model can call multiple tools in parallel."""
 
     prompt_mode: Annotated[
         OptionalNullable[MistralPromptMode], PlainValidator(validate_open_enum(False))

mistralai_azure/models/chatcompletionstreamrequest.py

@@ -59,11 +59,13 @@ ChatCompletionStreamRequestToolChoiceTypedDict = TypeAliasType(
     "ChatCompletionStreamRequestToolChoiceTypedDict",
     Union[ToolChoiceTypedDict, ToolChoiceEnum],
 )
+r"""Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool."""
 
 
 ChatCompletionStreamRequestToolChoice = TypeAliasType(
     "ChatCompletionStreamRequestToolChoice", Union[ToolChoice, ToolChoiceEnum]
 )
+r"""Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool."""
 
 
 class ChatCompletionStreamRequestTypedDict(TypedDict):
@@ -83,16 +85,21 @@ class ChatCompletionStreamRequestTypedDict(TypedDict):
     random_seed: NotRequired[Nullable[int]]
     r"""The seed to use for random sampling. If set, different calls will generate deterministic results."""
     response_format: NotRequired[ResponseFormatTypedDict]
+    r"""Specify the format that the model must output. By default it will use `{ \"type\": \"text\" }`. Setting to `{ \"type\": \"json_object\" }` enables JSON mode, which guarantees the message the model generates is in JSON. When using JSON mode you MUST also instruct the model to produce JSON yourself with a system or a user message. Setting to `{ \"type\": \"json_schema\" }` enables JSON schema mode, which guarantees the message the model generates is in JSON and follows the schema you provide."""
     tools: NotRequired[Nullable[List[ToolTypedDict]]]
+    r"""A list of tools the model may call. Use this to provide a list of functions the model may generate JSON inputs for."""
     tool_choice: NotRequired[ChatCompletionStreamRequestToolChoiceTypedDict]
+    r"""Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool."""
     presence_penalty: NotRequired[float]
-    r"""presence_penalty determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative."""
+    r"""The `presence_penalty` determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative."""
     frequency_penalty: NotRequired[float]
-    r"""frequency_penalty penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition."""
+    r"""The `frequency_penalty` penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition."""
     n: NotRequired[Nullable[int]]
     r"""Number of completions to return for each request, input tokens are only billed once."""
     prediction: NotRequired[PredictionTypedDict]
+    r"""Enable users to specify an expected completion, optimizing response times by leveraging known or predictable content."""
     parallel_tool_calls: NotRequired[bool]
+    r"""Whether to enable parallel function calling during tool use, when enabled the model can call multiple tools in parallel."""
     prompt_mode: NotRequired[Nullable[MistralPromptMode]]
     r"""Allows toggling between the reasoning mode and no system prompt. When set to `reasoning` the system prompt for reasoning models will be used."""
     safe_prompt: NotRequired[bool]
@@ -124,23 +131,28 @@ class ChatCompletionStreamRequest(BaseModel):
     r"""The seed to use for random sampling. If set, different calls will generate deterministic results."""
 
     response_format: Optional[ResponseFormat] = None
+    r"""Specify the format that the model must output. By default it will use `{ \"type\": \"text\" }`. Setting to `{ \"type\": \"json_object\" }` enables JSON mode, which guarantees the message the model generates is in JSON. When using JSON mode you MUST also instruct the model to produce JSON yourself with a system or a user message. Setting to `{ \"type\": \"json_schema\" }` enables JSON schema mode, which guarantees the message the model generates is in JSON and follows the schema you provide."""
 
     tools: OptionalNullable[List[Tool]] = UNSET
+    r"""A list of tools the model may call. Use this to provide a list of functions the model may generate JSON inputs for."""
 
     tool_choice: Optional[ChatCompletionStreamRequestToolChoice] = None
+    r"""Controls which (if any) tool is called by the model. `none` means the model will not call any tool and instead generates a message. `auto` means the model can pick between generating a message or calling one or more tools. `any` or `required` means the model must call one or more tools. Specifying a particular tool via `{\"type\": \"function\", \"function\": {\"name\": \"my_function\"}}` forces the model to call that tool."""
 
     presence_penalty: Optional[float] = None
-    r"""presence_penalty determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative."""
+    r"""The `presence_penalty` determines how much the model penalizes the repetition of words or phrases. A higher presence penalty encourages the model to use a wider variety of words and phrases, making the output more diverse and creative."""
 
     frequency_penalty: Optional[float] = None
-    r"""frequency_penalty penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition."""
+    r"""The `frequency_penalty` penalizes the repetition of words based on their frequency in the generated text. A higher frequency penalty discourages the model from repeating words that have already appeared frequently in the output, promoting diversity and reducing repetition."""
 
     n: OptionalNullable[int] = UNSET
     r"""Number of completions to return for each request, input tokens are only billed once."""
 
     prediction: Optional[Prediction] = None
+    r"""Enable users to specify an expected completion, optimizing response times by leveraging known or predictable content."""
 
     parallel_tool_calls: Optional[bool] = None
+    r"""Whether to enable parallel function calling during tool use, when enabled the model can call multiple tools in parallel."""
 
     prompt_mode: Annotated[
         OptionalNullable[MistralPromptMode], PlainValidator(validate_open_enum(False))