mirror of
https://github.com/meta-llama/llama-stack.git
synced 2025-07-14 17:16:09 +00:00
updated chat completion
This commit is contained in:
Hardik Shah
parent
749e50b288
commit
eb81ad1ffd
2 changed files with 177 additions and 122 deletions
|
|
@ -1,107 +1,13 @@
|
|||
openapi: 3.0.0
|
||||
info:
|
||||
title: Single Inference API (chat_completion)
|
||||
title: Chat Completion API
|
||||
version: 0.0.1
|
||||
components:
|
||||
schemas:
|
||||
Tool:
|
||||
type: object
|
||||
description: A tool that can be used by an agent to perform specific tasks.
|
||||
properties:
|
||||
name:
|
||||
type: string
|
||||
description: The name of the tool.
|
||||
description:
|
||||
type: string
|
||||
description: A brief description of what the tool does and how it should be used.
|
||||
parameters:
|
||||
type: array
|
||||
items:
|
||||
$ref: '#/components/schemas/ToolParameter'
|
||||
returnValue:
|
||||
$ref: '#/components/schemas/ToolReturnValue'
|
||||
ToolParameter:
|
||||
type: object
|
||||
properties:
|
||||
type:
|
||||
type: string
|
||||
enum: [string, int, float, list, bool]
|
||||
description: The data type of the parameter.
|
||||
itemType:
|
||||
type: string
|
||||
description: The type of items in the parameter if it is a list.
|
||||
description:
|
||||
type: string
|
||||
description: Details about what the parameter is used for and any constraints.
|
||||
ToolReturnValue:
|
||||
type: object
|
||||
properties:
|
||||
type:
|
||||
type: string
|
||||
enum: [string, int, float, list, bool]
|
||||
description: The data type of the return value.
|
||||
itemType:
|
||||
type: string
|
||||
description: The type of items in the return value if it is a list.
|
||||
description:
|
||||
type: string
|
||||
description: Details about the significance of the return value.
|
||||
Attachment:
|
||||
type: object
|
||||
properties:
|
||||
uri:
|
||||
type: string
|
||||
description: URI of the attachment.
|
||||
mime-type:
|
||||
type: string
|
||||
description: MIME type of the attachment.
|
||||
Message:
|
||||
type: object
|
||||
properties:
|
||||
role:
|
||||
type: string
|
||||
description: Role of the entity in the message.
|
||||
text:
|
||||
type: string
|
||||
description: Text content of the message.
|
||||
attachments:
|
||||
type: array
|
||||
items:
|
||||
$ref: '#/components/schemas/Attachment'
|
||||
Completion:
|
||||
type: object
|
||||
properties:
|
||||
id:
|
||||
type: string
|
||||
description: Unique identifier for the completion.
|
||||
role:
|
||||
type: string
|
||||
description: Role of the entity generating the completion.
|
||||
text:
|
||||
type: string
|
||||
description: Text content of the completion.
|
||||
attachments:
|
||||
type: array
|
||||
items:
|
||||
$ref: '#/components/schemas/Attachment'
|
||||
tokens:
|
||||
type: array
|
||||
items:
|
||||
type: integer
|
||||
logprobs:
|
||||
type: array
|
||||
items:
|
||||
type: number
|
||||
finish_reason:
|
||||
type: string
|
||||
description: Reason for completion termination.
|
||||
# TODO: Add `tool_choice` --
|
||||
# for eg. "auto": use model's guess, how to force to use particular tool, how to disbale inbuilt tools
|
||||
paths:
|
||||
/chat_completion/:
|
||||
post:
|
||||
summary: Submit a chat completion request
|
||||
description: Submit a chat completion request
|
||||
description: |
|
||||
This endpoint allows clients to submit a chat completion request.
|
||||
requestBody:
|
||||
required: true
|
||||
content:
|
||||
|
|
@ -115,31 +21,13 @@ paths:
|
|||
$ref: '#/components/schemas/Message'
|
||||
model:
|
||||
type: string
|
||||
description: Model identifier
|
||||
logprobs:
|
||||
type: boolean
|
||||
description: Whether to include log probabilities in the output
|
||||
max_tokens:
|
||||
type: integer
|
||||
description: Maximum number of tokens to generate
|
||||
options:
|
||||
$ref: '#/components/schemas/Options'
|
||||
n_completions:
|
||||
type: integer
|
||||
description: Number of completions to generate
|
||||
temperature:
|
||||
type: number
|
||||
format: float
|
||||
description: Temperature setting for the generation
|
||||
top_p:
|
||||
type: number
|
||||
format: float
|
||||
description: Top p setting for the generation
|
||||
tools:
|
||||
type: array
|
||||
items:
|
||||
$ref: '#/components/schemas/Tool'
|
||||
responses:
|
||||
'200':
|
||||
description: Chat completion request processed successfully
|
||||
description: Successful response
|
||||
content:
|
||||
application/json:
|
||||
schema:
|
||||
|
|
@ -147,11 +35,106 @@ paths:
|
|||
properties:
|
||||
id:
|
||||
type: string
|
||||
description: Unique identifier for the completion request
|
||||
completions:
|
||||
candidates:
|
||||
type: array
|
||||
items:
|
||||
$ref: '#/components/schemas/Completion'
|
||||
model:
|
||||
model_called:
|
||||
type: string
|
||||
description: Model used for generating completions
|
||||
usage:
|
||||
$ref: '#/components/schemas/TokenUsage'
|
||||
components:
|
||||
schemas:
|
||||
Message:
|
||||
type: object
|
||||
properties:
|
||||
role:
|
||||
type: string
|
||||
text:
|
||||
type: string
|
||||
attachments:
|
||||
type: array
|
||||
items:
|
||||
$ref: '#/components/schemas/MediaAttachment'
|
||||
eot:
|
||||
type: boolean
|
||||
description: "End of transmission flag."
|
||||
tool_call:
|
||||
type: boolean
|
||||
description: "Indicates if it's a tool call - builtin, custom, or ipython."
|
||||
is_complete:
|
||||
type: boolean
|
||||
description: "For streaming, indicates if the message is complete."
|
||||
is_header_complete:
|
||||
type: boolean
|
||||
description: "For streaming, indicates if the header of the message is complete."
|
||||
metadata:
|
||||
type: object
|
||||
additionalProperties: true
|
||||
description: "Additional metadata as JSON."
|
||||
MediaAttachment:
|
||||
type: object
|
||||
properties:
|
||||
attachment_type:
|
||||
$ref: '#/components/schemas/MediaAttachmentType'
|
||||
data_type:
|
||||
$ref: '#/components/schemas/MediaAttachmentDataType'
|
||||
data:
|
||||
type: string
|
||||
MediaAttachmentType:
|
||||
type: string
|
||||
enum:
|
||||
- image
|
||||
- video
|
||||
- audio
|
||||
- text
|
||||
description: "Type of media attachment."
|
||||
MediaAttachmentDataType:
|
||||
type: string
|
||||
enum:
|
||||
- raw_bytes
|
||||
- filepath
|
||||
- uri
|
||||
description: "Data type of the media attachment."
|
||||
Completion:
|
||||
type: object
|
||||
properties:
|
||||
id:
|
||||
type: string
|
||||
message:
|
||||
$ref: '#/components/schemas/Message'
|
||||
tokens:
|
||||
type: array
|
||||
items:
|
||||
type: integer
|
||||
logprobs:
|
||||
type: array
|
||||
items:
|
||||
type: number
|
||||
finish_reason:
|
||||
type: string
|
||||
enum:
|
||||
- stop
|
||||
- safety
|
||||
- max-length
|
||||
description: "Reason for completion termination."
|
||||
Options:
|
||||
type: object
|
||||
properties:
|
||||
logprobs:
|
||||
type: boolean
|
||||
max_tokens:
|
||||
type: integer
|
||||
temperature:
|
||||
type: number
|
||||
top_p:
|
||||
type: number
|
||||
TokenUsage:
|
||||
type: object
|
||||
properties:
|
||||
input_tokens:
|
||||
type: integer
|
||||
output_tokens:
|
||||
type: integer
|
||||
total_tokens:
|
||||
type: integer
|
||||
|
|
|
|||
72
simple_view/chat_completion.yml
Normal file
72
simple_view/chat_completion.yml
Normal file
|
|
@ -0,0 +1,72 @@
|
|||
# Simple bullet form for ease of read and iteration
|
||||
# Use LLMs to translate this to a OpenAPI spec.
|
||||
|
||||
== Schema ==
|
||||
|
||||
Message:
|
||||
role: str
|
||||
text: str
|
||||
attachements: List[MediaAttachment]
|
||||
eot: bool
|
||||
tool_call: bool # if it's a tool call - builtin or custom or ipython
|
||||
# for streaming
|
||||
is_complete: bool
|
||||
is_header_complete: bool
|
||||
metadata: json
|
||||
|
||||
MediaAttachment:
|
||||
attachement_type: MediaAttachmentType
|
||||
data_type: MediaAttachmentDataType
|
||||
data: str
|
||||
|
||||
MediaAttachmentType: # enum [image, video, audio, text(or file)]
|
||||
MediaAttachmentDataType: # enum [raw_bytes, filepath, uri]
|
||||
|
||||
Completion:
|
||||
id: str
|
||||
message: Message
|
||||
tokens: List[int]
|
||||
logprobs: List[floats]
|
||||
finish_reason: str # Enum (stop, safety, max-length, etc)
|
||||
|
||||
Options:
|
||||
logprobs: bool
|
||||
max_tokens: int
|
||||
temperature: float
|
||||
top_p: float
|
||||
#TODO: Get more options from metagen
|
||||
|
||||
TokenUsage:
|
||||
input_tokens: int
|
||||
output_tokens: int
|
||||
total_tokens: int
|
||||
|
||||
== Callsite ==
|
||||
|
||||
callsite:
|
||||
chat_completion/
|
||||
request_type:
|
||||
post
|
||||
description:
|
||||
submit a chat completion request
|
||||
request:
|
||||
messages: List[Message]
|
||||
model: str
|
||||
options: Options
|
||||
n_complections: int
|
||||
# TODO: how to handle tooling control if any ?
|
||||
# Add `tools` and `tool_choice` --
|
||||
# for eg. "auto": use model's guess
|
||||
# how to force to use particular tool
|
||||
# how to disbale inbuilt tools
|
||||
# tools: List[Tool]
|
||||
# tool_choice: Any
|
||||
response:
|
||||
id: str
|
||||
candidates: List[Completion] # a list to account for when n_completions > 1
|
||||
model_called: str # info on that model that produced this result
|
||||
usage: TokenUsage
|
||||
|
||||
# TODO
|
||||
# callsite:
|
||||
# chat_completion_stream/
|
||||
Loading…
Add table
Add a link
Reference in a new issue