feat(api): OpenAPI spec update via Stainless API

.stats.yml

@@ -1,2 +1,2 @@
 configured_endpoints: 15
-openapi_spec_url: https://storage.googleapis.com/stainless-sdk-openapi-specs/togetherai%2FTogetherAI-33661dd8fd4c26ecd595dee22e2c9274e6c4699ad8de5ece233e0d37376c6b7c.yml
+openapi_spec_url: https://storage.googleapis.com/stainless-sdk-openapi-specs/togetherai%2FTogetherAI-e8f4e11a2e3927c75dce42c913ef5c9adcf2aef3d3b1312b4825d9f135413c39.yml

README.md

@@ -362,7 +362,7 @@ You can directly override the [httpx client](https://www.python-httpx.org/api/#c
 
 - Support for proxies
 - Custom transports
-- Additional [advanced](https://www.python-httpx.org/advanced/#client-instances) functionality
+- Additional [advanced](https://www.python-httpx.org/advanced/clients/) functionality
 
 ```python
 from together import Together, DefaultHttpxClient

bin/publish-pypi

@@ -3,4 +3,7 @@
 set -eux
 mkdir -p dist
 rye build --clean
+# Patching importlib-metadata version until upstream library version is updated
+# https://github.com/pypa/twine/issues/977#issuecomment-2189800841
+"$HOME/.rye/self/bin/python3" -m pip install 'importlib-metadata==7.2.1'
 rye publish --yes --token=$PYPI_TOKEN

src/together/resources/chat/completions.py

@@ -50,6 +50,7 @@ def create(
         model: str,
         echo: bool | NotGiven = NOT_GIVEN,
         frequency_penalty: float | NotGiven = NOT_GIVEN,
+        function_call: completion_create_params.FunctionCall | NotGiven = NOT_GIVEN,
         logit_bias: Dict[str, float] | NotGiven = NOT_GIVEN,
         logprobs: int | NotGiven = NOT_GIVEN,
         max_tokens: int | NotGiven = NOT_GIVEN,
@@ -161,6 +162,7 @@ def create(
         stream: Literal[True],
         echo: bool | NotGiven = NOT_GIVEN,
         frequency_penalty: float | NotGiven = NOT_GIVEN,
+        function_call: completion_create_params.FunctionCall | NotGiven = NOT_GIVEN,
         logit_bias: Dict[str, float] | NotGiven = NOT_GIVEN,
         logprobs: int | NotGiven = NOT_GIVEN,
         max_tokens: int | NotGiven = NOT_GIVEN,
@@ -271,6 +273,7 @@ def create(
         stream: bool,
         echo: bool | NotGiven = NOT_GIVEN,
         frequency_penalty: float | NotGiven = NOT_GIVEN,
+        function_call: completion_create_params.FunctionCall | NotGiven = NOT_GIVEN,
         logit_bias: Dict[str, float] | NotGiven = NOT_GIVEN,
         logprobs: int | NotGiven = NOT_GIVEN,
         max_tokens: int | NotGiven = NOT_GIVEN,
@@ -380,6 +383,7 @@ def create(
         model: str,
         echo: bool | NotGiven = NOT_GIVEN,
         frequency_penalty: float | NotGiven = NOT_GIVEN,
+        function_call: completion_create_params.FunctionCall | NotGiven = NOT_GIVEN,
         logit_bias: Dict[str, float] | NotGiven = NOT_GIVEN,
         logprobs: int | NotGiven = NOT_GIVEN,
         max_tokens: int | NotGiven = NOT_GIVEN,
@@ -411,6 +415,7 @@ def create(
                     "model": model,
                     "echo": echo,
                     "frequency_penalty": frequency_penalty,
+                    "function_call": function_call,
                     "logit_bias": logit_bias,
                     "logprobs": logprobs,
                     "max_tokens": max_tokens,
@@ -456,6 +461,7 @@ async def create(
         model: str,
         echo: bool | NotGiven = NOT_GIVEN,
         frequency_penalty: float | NotGiven = NOT_GIVEN,
+        function_call: completion_create_params.FunctionCall | NotGiven = NOT_GIVEN,
         logit_bias: Dict[str, float] | NotGiven = NOT_GIVEN,
         logprobs: int | NotGiven = NOT_GIVEN,
         max_tokens: int | NotGiven = NOT_GIVEN,
@@ -567,6 +573,7 @@ async def create(
         stream: Literal[True],
         echo: bool | NotGiven = NOT_GIVEN,
         frequency_penalty: float | NotGiven = NOT_GIVEN,
+        function_call: completion_create_params.FunctionCall | NotGiven = NOT_GIVEN,
         logit_bias: Dict[str, float] | NotGiven = NOT_GIVEN,
         logprobs: int | NotGiven = NOT_GIVEN,
         max_tokens: int | NotGiven = NOT_GIVEN,
@@ -677,6 +684,7 @@ async def create(
         stream: bool,
         echo: bool | NotGiven = NOT_GIVEN,
         frequency_penalty: float | NotGiven = NOT_GIVEN,
+        function_call: completion_create_params.FunctionCall | NotGiven = NOT_GIVEN,
         logit_bias: Dict[str, float] | NotGiven = NOT_GIVEN,
         logprobs: int | NotGiven = NOT_GIVEN,
         max_tokens: int | NotGiven = NOT_GIVEN,
@@ -786,6 +794,7 @@ async def create(
         model: str,
         echo: bool | NotGiven = NOT_GIVEN,
         frequency_penalty: float | NotGiven = NOT_GIVEN,
+        function_call: completion_create_params.FunctionCall | NotGiven = NOT_GIVEN,
         logit_bias: Dict[str, float] | NotGiven = NOT_GIVEN,
         logprobs: int | NotGiven = NOT_GIVEN,
         max_tokens: int | NotGiven = NOT_GIVEN,
@@ -817,6 +826,7 @@ async def create(
                     "model": model,
                     "echo": echo,
                     "frequency_penalty": frequency_penalty,
+                    "function_call": function_call,
                     "logit_bias": logit_bias,
                     "logprobs": logprobs,
                     "max_tokens": max_tokens,

src/together/types/__init__.py

@@ -7,6 +7,7 @@
 from .log_probs import LogProbs as LogProbs
 from .completion import Completion as Completion
 from .image_file import ImageFile as ImageFile
+from .tool_choice import ToolChoice as ToolChoice
 from .tools_param import ToolsParam as ToolsParam
 from .fine_tune_event import FineTuneEvent as FineTuneEvent
 from .tool_choice_param import ToolChoiceParam as ToolChoiceParam

src/together/types/chat/chat_completion.py

@@ -5,34 +5,51 @@
 
 from ..._models import BaseModel
 from ..log_probs import LogProbs
+from ..tool_choice import ToolChoice
 from .chat_completion_usage import ChatCompletionUsage
 
-__all__ = ["ChatCompletion", "Choice", "ChoiceMessage"]
+__all__ = ["ChatCompletion", "Choice", "ChoiceMessage", "ChoiceMessageFunctionCall"]
+
+
+class ChoiceMessageFunctionCall(BaseModel):
+    arguments: str
+
+    name: str
 
 
 class ChoiceMessage(BaseModel):
     content: Optional[str] = None
 
-    role: Optional[str] = None
+    role: Literal["assistant"]
+
+    function_call: Optional[ChoiceMessageFunctionCall] = None
+
+    tool_calls: Optional[List[ToolChoice]] = None
 
 
 class Choice(BaseModel):
-    finish_reason: Optional[Literal["stop", "eos", "length", "tool_calls"]] = None
+    finish_reason: Optional[Literal["stop", "eos", "length", "tool_calls", "function_call"]] = None
+
+    index: Optional[int] = None
 
     logprobs: Optional[LogProbs] = None
 
     message: Optional[ChoiceMessage] = None
 
+    seed: Optional[int] = None
+
+    text: Optional[str] = None
+
 
 class ChatCompletion(BaseModel):
-    id: Optional[str] = None
+    id: str
 
-    choices: Optional[List[Choice]] = None
+    choices: List[Choice]
 
-    created: Optional[int] = None
+    created: int
 
-    model: Optional[str] = None
+    model: str
 
-    object: Optional[Literal["chat.completion"]] = None
+    object: Literal["chat.completion"]
 
     usage: Optional[ChatCompletionUsage] = None

src/together/types/chat/chat_completion_chunk.py

@@ -4,42 +4,52 @@
 from typing_extensions import Literal
 
 from ..._models import BaseModel
+from ..log_probs import LogProbs
+from ..tool_choice import ToolChoice
 from .chat_completion_usage import ChatCompletionUsage
 
-__all__ = ["ChatCompletionChunk", "Token", "Choice", "ChoiceDelta"]
+__all__ = ["ChatCompletionChunk", "Choice", "ChoiceDelta", "ChoiceDeltaFunctionCall"]
 
 
-class Token(BaseModel):
-    id: int
+class ChoiceDeltaFunctionCall(BaseModel):
+    arguments: str
 
-    logprob: float
+    name: str
 
-    special: bool
 
-    text: str
+class ChoiceDelta(BaseModel):
+    content: Optional[str] = None
 
+    function_call: Optional[ChoiceDeltaFunctionCall] = None
 
-class ChoiceDelta(BaseModel):
-    content: str
+    role: Optional[Literal["system", "user", "assistant", "function", "tool"]] = None
+
+    token_id: Optional[int] = None
+
+    tool_calls: Optional[List[ToolChoice]] = None
 
 
 class Choice(BaseModel):
     delta: ChoiceDelta
 
+    finish_reason: Literal["stop", "eos", "length", "tool_calls", "function_call"]
+
     index: int
 
+    logprobs: Optional[LogProbs] = None
+
 
 class ChatCompletionChunk(BaseModel):
     id: str
 
-    token: Token
-
     choices: List[Choice]
 
     created: int
 
+    model: str
+
     object: Literal["chat.completion.chunk"]
 
-    finish_reason: Optional[Literal["stop", "eos", "length", "tool_calls"]] = None
+    system_fingerprint: Optional[str] = None
 
     usage: Optional[ChatCompletionUsage] = None

src/together/types/chat/completion_create_params.py

@@ -11,6 +11,8 @@
 __all__ = [
     "CompletionCreateParamsBase",
     "Message",
+    "FunctionCall",
+    "FunctionCallName",
     "ResponseFormat",
     "ToolChoice",
     "CompletionCreateParamsNonStreaming",
@@ -37,6 +39,8 @@ class CompletionCreateParamsBase(TypedDict, total=False):
     repeating tokens that have already been mentioned.
     """
 
+    function_call: FunctionCall
+
     logit_bias: Dict[str, float]
     """Adjusts the likelihood of specific tokens appearing in the generated output."""
 
@@ -133,6 +137,13 @@ class Message(TypedDict, total=False):
     """The role of the messages author. Choice between: system, user, or assistant."""
 
 
+class FunctionCallName(TypedDict, total=False):
+    name: Required[str]
+
+
+FunctionCall = Union[Literal["none", "auto"], FunctionCallName]
+
+
 class ResponseFormat(TypedDict, total=False):
     schema: Dict[str, str]
     """The schema of the response format."""

src/together/types/completion.py

@@ -11,7 +11,7 @@
 
 
 class Choice(BaseModel):
-    finish_reason: Optional[Literal["stop", "eos", "length", "tool_calls"]] = None
+    finish_reason: Optional[Literal["stop", "eos", "length", "tool_calls", "function_call"]] = None
 
     logprobs: Optional[LogProbs] = None
 

src/together/types/fine_tune_event.py

@@ -1,6 +1,5 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
-import builtins
 from typing import List, Optional
 from typing_extensions import Literal
 
@@ -64,7 +63,7 @@ class Data(BaseModel):
 
     wandb_url: str
 
-    level: Optional[builtins.object] = None
+    level: Optional[Literal["info", "warning", "error", "legacy_info", "legacy_iwarning", "legacy_ierror"]] = None
 
 
 class FineTuneEvent(BaseModel):

src/together/types/log_probs.py

@@ -4,10 +4,18 @@
 
 from .._models import BaseModel
 
-__all__ = ["LogProbs"]
+__all__ = ["LogProbs", "Content"]
+
+
+class Content(BaseModel):
+    token: str
+
+    logprob: float
 
 
 class LogProbs(BaseModel):
+    content: Optional[List[Content]] = None
+
     token_logprobs: Optional[List[float]] = None
     """List of token log probabilities"""
 

src/together/types/tool_choice.py

@@ -0,0 +1,23 @@
+# File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
+
+from typing_extensions import Literal
+
+from .._models import BaseModel
+
+__all__ = ["ToolChoice", "Function"]
+
+
+class Function(BaseModel):
+    arguments: str
+
+    name: str
+
+
+class ToolChoice(BaseModel):
+    id: str
+
+    function: Function
+
+    index: float
+
+    type: Literal["function"]

src/together/types/tool_choice_param.py

@@ -2,16 +2,22 @@
 
 from __future__ import annotations
 
-from typing_extensions import TypedDict
+from typing_extensions import Literal, Required, TypedDict
 
 __all__ = ["ToolChoiceParam", "Function"]
 
 
 class Function(TypedDict, total=False):
-    name: str
+    arguments: Required[str]
+
+    name: Required[str]
 
 
 class ToolChoiceParam(TypedDict, total=False):
-    function: Function
+    id: Required[str]
+
+    function: Required[Function]
+
+    index: Required[float]
 
-    type: str
+    type: Required[Literal["function"]]

tests/api_resources/chat/test_completions.py

@@ -58,6 +58,7 @@ def test_method_create_with_all_params_overload_1(self, client: Together) -> Non
             model="mistralai/Mixtral-8x7B-Instruct-v0.1",
             echo=True,
             frequency_penalty=0,
+            function_call="none",
             logit_bias={
                 "105": 21.4,
                 "1024": -10.5,
@@ -203,6 +204,7 @@ def test_method_create_with_all_params_overload_2(self, client: Together) -> Non
             stream=True,
             echo=True,
             frequency_penalty=0,
+            function_call="none",
             logit_bias={
                 "105": 21.4,
                 "1024": -10.5,
@@ -350,6 +352,7 @@ async def test_method_create_with_all_params_overload_1(self, async_client: Asyn
             model="mistralai/Mixtral-8x7B-Instruct-v0.1",
             echo=True,
             frequency_penalty=0,
+            function_call="none",
             logit_bias={
                 "105": 21.4,
                 "1024": -10.5,
@@ -495,6 +498,7 @@ async def test_method_create_with_all_params_overload_2(self, async_client: Asyn
             stream=True,
             echo=True,
             frequency_penalty=0,
+            function_call="none",
             logit_bias={
                 "105": 21.4,
                 "1024": -10.5,