Reference: Chat Completions | Azure AI Foundry

Artikkeli
08/28/2024

Important

Items marked (preview) in this article are currently in public preview. This preview is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews.

Creates a model response for the given chat conversation.

POST /chat/completions?api-version=2024-04-01-preview

URI Parameters

Name	In	Required	Type	Description
api-version	query	True	string	The version of the API in the format "YYYY-MM-DD" or "YYYY-MM-DD-preview".

Request Header

Name	Required	Type	Description
extra-parameters		string	The behavior of the API when extra parameters are indicated in the payload. Using `pass-through` makes the API to pass the parameter to the underlying model. Use this value when you want to pass parameters that you know the underlying model can support. Using `ignore` makes the API to drop any unsupported parameter. Use this value when you need to use the same payload across different models, but one of the extra parameters may make a model to error out if not supported. Using `error` makes the API to reject any extra parameter in the payload. Only parameters specified in this API can be indicated, or a 400 error is returned.
azureml-model-deployment		string	Name of the deployment you want to route the request to. Supported for endpoints that support multiple deployments.

Request Body

Name	Required	Type	Description
model		string	The model name. This parameter is ignored if the endpoint serves only one model.
messages	True	ChatCompletionRequestMessage	A list of messages comprising the conversation so far. Returns a 422 error if at least some of the messages can't be understood by the model.
frequency_penalty		number	Helps prevent word repetitions by reducing the chance of a word being selected if it has already been used. The higher the frequency penalty, the less likely the model is to repeat the same words in its output. Return a 422 error if value or parameter is not supported by model.
max_tokens		integer	The maximum number of tokens that can be generated in the chat completion. The total length of input tokens and generated tokens is limited by the model's context length. Passing null causes the model to use its max context length.
presence_penalty		number	Helps prevent the same topics from being repeated by penalizing a word if it exists in the completion already, even just once. Return a 422 error if value or parameter is not supported by model.
response_format		ChatCompletionResponseFormat
seed		integer	If specified, our system will make a best effort to sample deterministically, such that repeated requests with the same `seed` and parameters should return the same result. Determinism is not guaranteed, and you should refer to the `system_fingerprint` response parameter to monitor changes in the backend.
stop			Sequences where the API will stop generating further tokens.
stream		boolean	If set, partial message deltas will be sent. Tokens will be sent as data-only server-sent events as they become available, with the stream terminated by a `data: [DONE]` message.
temperature		number	Non-negative number. Return 422 if value is unsupported by model.
tool_choice		ChatCompletionToolChoiceOption	Controls which (if any) function is called by the model. `none` means the model will not call a function and instead generates a message. `auto` means the model can pick between generating a message or calling a function. Specifying a particular function via `{"type": "function", "function": {"name": "my_function"}}` forces the model to call that function. `none` is the default when no functions are present. `auto` is the default if functions are present. Returns a 422 error if the tool is not supported by the model.
tools		ChatCompletionTool[]	A list of tools the model may call. Currently, only functions are supported as a tool. Use this to provide a list of functions the model may generate JSON inputs for. Returns a 422 error if the tool is not supported by the model.
top_p		number	An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. We generally recommend altering this or `temperature` but not both.

Responses

Name	Type	Description
200 OK	CreateChatCompletionResponse	OK
401 Unauthorized	UnauthorizedError	Access token is missing or invalid Headers x-ms-error-code: string
404 Not Found	NotFoundError	Modality not supported by the model. Check the documentation of the model to see which routes are available. Headers x-ms-error-code: string
422 Unprocessable Entity	UnprocessableContentError	The request contains unprocessable content Headers x-ms-error-code: string
429 Too Many Requests	TooManyRequestsError	You have hit your assigned rate limit and your request need to be paced. Headers x-ms-error-code: string
Other Status Codes	ContentFilterError	Bad request Headers x-ms-error-code: string

Security

Authorization

The token with the Bearer: prefix, e.g. Bearer abcde12345

Type: apiKey
In: header

AADToken

Azure Active Directory OAuth2 authentication

Type: oauth2
Flow: application
Token URL: https://login.microsoftonline.com/common/oauth2/v2.0/token

Examples

Creates a model response for the given chat conversation

Sample Request

POST /chat/completions?api-version=2024-04-01-preview

{
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant"
    },
    {
      "role": "user",
      "content": "Explain Riemann's conjecture"
    },
    {
      "role": "assistant",
      "content": "The Riemann Conjecture is a deep mathematical conjecture around prime numbers and how they can be predicted. It was first published in Riemann's groundbreaking 1859 paper. The conjecture states that the Riemann zeta function has its zeros only at the negative even integers and complex numbers with real part 1/21. Many consider it to be the most important unsolved problem in pure mathematics. The Riemann hypothesis is a way to predict the probability that numbers in a certain range are prime that was also devised by German mathematician Bernhard Riemann in 18594."
    },
    {
      "role": "user",
      "content": "Ist it proved?"
    }
  ],
  "frequency_penalty": 0,
  "presence_penalty": 0,
  "max_tokens": 256,
  "seed": 42,
  "stop": "<|endoftext|>",
  "stream": false,
  "temperature": 0,
  "top_p": 1,
  "response_format": { "type": "text" }
}

Sample Response

Status code: 200

{
  "id": "1234567890",
  "model": "llama2-70b-chat",
  "choices": [
    {
      "index": 0,
      "finish_reason": "stop",
      "message": {
        "role": "assistant",
        "content": "No, it has never been proved"
      }
    }
  ],
  "created": 1234567890,
  "object": "chat.completion",
  "usage": {
    "prompt_tokens": 205,
    "completion_tokens": 5,
    "total_tokens": 210
  }
}

Definitions

Name	Description
ChatCompletionRequestMessage
ChatCompletionMessageContentPart
ChatCompletionMessageContentPartType
ChatCompletionToolChoiceOption	Controls which (if any) function is called by the model. `none` means the model will not call a function and instead generates a message. `auto` means the model can pick between generating a message or calling a function. Specifying a particular function via `{"type": "function", "function": {"name": "my_function"}}` forces the model to call that function. `none` is the default when no functions are present. `auto` is the default if functions are present. Returns a 422 error if the tool is not supported by the model.
ChatCompletionFinishReason	The reason the model stopped generating tokens. This will be `stop` if the model hit a natural stop point or a provided stop sequence, `length` if the maximum number of tokens specified in the request was reached, `content_filter` if content was omitted due to a flag from our content filters, `tool_calls` if the model called a tool.
ChatCompletionMessageToolCall
ChatCompletionObject	The object type, which is always `chat.completion`.
ChatCompletionResponseFormat	The response format for the model response. Setting to `json_object` enables JSON mode, which guarantees the message the model generates is valid JSON. When using JSON mode, you must also instruct the model to produce JSON yourself via a system or user message. Also note that the message content may be partially cut off if `finish_reason="length"`, which indicates the generation exceeded `max_tokens` or the conversation exceeded the max context length.
ChatCompletionResponseFormatType	The response format type.
ChatCompletionResponseMessage	A chat completion message generated by the model.
ChatCompletionTool
ChatMessageRole	The role of the author of this message.
Choices	A list of chat completion choices.
CompletionUsage	Usage statistics for the completion request.
ContentFilterError	The API call fails when the prompt triggers a content filter as configured. Modify the prompt and try again.
CreateChatCompletionRequest
CreateChatCompletionResponse	Represents a chat completion response returned by model, based on the provided input.
Detail	Details for the UnprocessableContentError error.
Function	The function that the model called.
FunctionObject	Definition of a function the model has access to.
ImageDetail	Specifies the detail level of the image.
NotFoundError	The route is not valid for the deployed model.
ToolType	The type of the tool. Currently, only `function` is supported.
TooManyRequestsError	You have hit your assigned rate limit and your requests need to be paced.
UnauthorizedError	Authentication is missing or invalid.
UnprocessableContentError	The request contains unprocessable content. The error is returned when the payload indicated is valid according to this specification. However, some of the instructions indicated in the payload are not supported by the underlying model. Use the `details` section to understand the offending parameter.

ChatCompletionFinishReason

The reason the model stopped generating tokens. This will be stop if the model hit a natural stop point or a provided stop sequence, length if the maximum number of tokens specified in the request was reached, content_filter if content was omitted due to a flag from our content filters, tool_calls if the model called a tool.

Name	Type	Description
content_filter	string
length	string
stop	string
tool_calls	string

ChatCompletionMessageToolCall

Name	Type	Description
function	Function	The function that the model called.
ID	string	The ID of the tool call.
type	ToolType	The type of the tool. Currently, only `function` is supported.

ChatCompletionObject

The object type, which is always chat.completion.

Name	Type	Description
chat.completion	string

ChatCompletionResponseFormat

The response format for the model response. Setting to json_object enables JSON mode, which guarantees the message the model generates is valid JSON. When using JSON mode, you must also instruct the model to produce JSON yourself via a system or user message. Also note that the message content may be partially cut off if finish_reason="length", which indicates the generation exceeded max_tokens or the conversation exceeded the max context length.

Name	Type	Description
type	ChatCompletionResponseFormatType	The response format type.

ChatCompletionResponseFormatType

The response format type.

Name	Type	Description
json_object	string
text	string

ChatCompletionResponseMessage

A chat completion message generated by the model.

Name	Type	Description
content	string	The contents of the message.
role	ChatMessageRole	The role of the author of this message.
tool_calls	ChatCompletionMessageToolCall[]	The tool calls generated by the model, such as function calls.

ChatCompletionTool

Name	Type	Description
function	FunctionObject
type	ToolType	The type of the tool. Currently, only `function` is supported.

ChatMessageRole

The role of the author of this message.

Name	Type	Description
assistant	string
system	string
tool	string
user	string

Choices

A list of chat completion choices. Can be more than one if n is greater than 1.

Name	Type	Description
finish_reason	ChatCompletionFinishReason	The reason the model stopped generating tokens. This will be `stop` if the model hit a natural stop point or a provided stop sequence, `length` if the maximum number of tokens specified in the request was reached, `content_filter` if content was omitted due to a flag from our content filters, `tool_calls` if the model called a tool.
index	integer	The index of the choice in the list of choices.
message	ChatCompletionResponseMessage	A chat completion message generated by the model.

CompletionUsage

Usage statistics for the completion request.

Name	Type	Description
completion_tokens	integer	Number of tokens in the generated completion.
prompt_tokens	integer	Number of tokens in the prompt.
total_tokens	integer	Total number of tokens used in the request (prompt + completion).

ContentFilterError

The API call fails when the prompt triggers a content filter as configured. Modify the prompt and try again.

Name	Type	Description
code	string	The error code.
error	string	The error description.
message	string	The error message.
param	string	The parameter that triggered the content filter.
status	integer	The HTTP status code.

CreateChatCompletionRequest

Name	Type	Default Value	Description
frequency_penalty	number	0	Helps prevent word repetitions by reducing the chance of a word being selected if it has already been used. The higher the frequency penalty, the less likely the model is to repeat the same words in its output. Return a 422 error if value or parameter is not supported by model.
max_tokens	integer		The maximum number of tokens that can be generated in the chat completion. The total length of input tokens and generated tokens is limited by the model's context length. Passing null causes the model to use its max context length.
messages	ChatCompletionRequestMessage[]		A list of messages comprising the conversation so far. Returns a 422 error if at least some of the messages can't be understood by the model.
presence_penalty	number	0	Helps prevent the same topics from being repeated by penalizing a word if it exists in the completion already, even just once. Return a 422 error if value or parameter is not supported by model.
response_format	ChatCompletionResponseFormat	text
seed	integer		If specified, our system will make a best effort to sample deterministically, such that repeated requests with the same `seed` and parameters should return the same result. Determinism is not guaranteed, and you should refer to the `system_fingerprint` response parameter to monitor changes in the backend.
stop			Sequences where the API will stop generating further tokens.
stream	boolean	False	If set, partial message deltas will be sent. Tokens will be sent as data-only server-sent events as they become available, with the stream terminated by a `data: [DONE]` message.
temperature	number	1	Non-negative number. Return 422 if value is unsupported by model.
tool_choice	ChatCompletionToolChoiceOption		Controls which (if any) function is called by the model. `none` means the model will not call a function and instead generates a message. `auto` means the model can pick between generating a message or calling a function. Specifying a particular function via `{"type": "function", "function": {"name": "my_function"}}` forces the model to call that function. `none` is the default when no functions are present. `auto` is the default if functions are present. Returns a 422 error if the tool is not supported by the model.
tools	ChatCompletionTool[]		A list of tools the model may call. Currently, only functions are supported as a tool. Use this to provide a list of functions the model may generate JSON inputs for. Returns a 422 error if the tool is not supported by the model.
top_p	number	1	An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. We generally recommend altering this or `temperature` but not both.

ChatCompletionRequestMessage

Name	Type	Description
content	string or ChatCompletionMessageContentPart[]	The contents of the message.
role	ChatMessageRole	The role of the author of this message.
tool_calls	ChatCompletionMessageToolCall[]	The tool calls generated by the model, such as function calls.

ChatCompletionMessageContentPart

Name	Type	Description
content	string	Either a URL of the image or the base64 encoded image data.
detail	ImageDetail	Specifies the detail level of the image.
type	ChatCompletionMessageContentPartType	The type of the content part.

ChatCompletionMessageContentPartType

Name	Type	Description
text	string
image	string
image_url	string

ChatCompletionToolChoiceOption

Controls which (if any) tool is called by the model.

Name	Type	Description
none	string	The model will not call any tool and instead generates a message.
auto	string	The model can pick between generating a message or calling one or more tools.
required	string	The model must call one or more tools.
	string	Specifying a particular tool via `{"type": "function", "function": {"name": "my_function"}}` forces the model to call that tool.

ImageDetail

Specifies the detail level of the image.

Name	Type	Description
auto	string
low	string
high	string

CreateChatCompletionResponse

Represents a chat completion response returned by model, based on the provided input.

Name	Type	Description
choices	Choices[]	A list of chat completion choices. Can be more than one if `n` is greater than 1.
created	integer	The Unix timestamp (in seconds) of when the chat completion was created.
ID	string	A unique identifier for the chat completion.
model	string	The model used for the chat completion.
object	ChatCompletionObject	The object type, which is always `chat.completion`.
system_fingerprint	string	This fingerprint represents the backend configuration that the model runs with. Can be used in conjunction with the `seed` request parameter to understand when backend changes have been made that might impact determinism.
usage	CompletionUsage	Usage statistics for the completion request.

Detail

Details for the UnprocessableContentError error.

Name	Type	Description
loc	string[]	The parameter causing the issue
value	string	The value passed to the parameter causing issues.

Function

The function that the model called.

Name	Type	Description
arguments	string	The arguments to call the function with, as generated by the model in JSON format. Note that the model does not always generate valid JSON, and may generate incorrect parameters not defined by your function schema. Validate the arguments in your code before calling your function.
name	string	The name of the function to call.

FunctionObject

Definition of a function the model has access to.

Name	Type	Description
description	string	A description of what the function does, used by the model to choose when and how to call the function.
name	string	The name of the function to be called. Must be a-z, A-Z, 0-9, or contain underscores and dashes, with a maximum length of 64.
parameters	object	The parameters the functions accepts, described as a JSON Schema object. Omitting `parameters` defines a function with an empty parameter list.

NotFoundError

Name	Type	Description
error	string	The error description.
message	string	The error message.
status	integer	The HTTP status code.

ToolType

The type of the tool. Currently, only function is supported.

Name	Type	Description
function	string

TooManyRequestsError

Name	Type	Description
error	string	The error description.
message	string	The error message.
status	integer	The HTTP status code.

UnauthorizedError

Name	Type	Description
error	string	The error description.
message	string	The error message.
status	integer	The HTTP status code.

UnprocessableContentError

The request contains unprocessable content. The error is returned when the payload indicated is valid according to this specification. However, some of the instructions indicated in the payload are not supported by the underlying model. Use the details section to understand the offending parameter.

Name	Type	Description
code	string	The error code.
detail	Detail
error	string	The error description.
message	string	The error message.
status	integer	The HTTP status code.

Jaa