LLMModelCreate
Payload used to create LLM Model. Required fields are name, type, provider, model.
- Option 1
- Option 2
Possible values: >= 2 characters and <= 128 characters, Value must match regular expression ^[a-z][a-z0-9_.]{1,127}(?![\s\S])
alias object
- string
- null
Model placement: internal hosts the model in D.Hub; external registers an existing endpoint.
Possible values: [internal, external]
Provider identifier returned by the provider-list operation.
label object
- string
- null
hint_i18n_key object
- string
- null
Possible values: non-empty
context_window object
Maximum model context size in tokens.
- integer
- null
Possible values: >= 1
tokenizer object
Tokenizer identifier used for token counting, given as a Hugging Face repository id. Omit this field to use the built-in estimator.
- string
- null
Possible values: non-empty and <= 256 characters, Value must match regular expression ^[A-Za-z0-9][A-Za-z0-9._-]*(/[A-Za-z0-9][A-Za-z0-9._-]*)*$
deployment object
- LLMModelDeployment
- null
Fully qualified container image reference for the model server.
command object
Optional container entrypoint override.
- string[]
- null
options object
Runtime options converted to command-line flags when the model starts.
property name* object
- string
- integer
- number
- boolean
- null
args object
Additional command-line arguments appended after rendered runtime options.
- string[]
- null
8080cpu object
Requested CPU amount, such as 500m.
- string
- null
memory object
Requested memory amount, such as 2Gi.
- string
- null
gpu object
GPU count request
- integer
- null
1/healthzenv object
Non-secret environment variables passed to the model container.
node_selector object
base_url object
- string
- null
api_key object
- string
- null
default_headers object
Custom default header rows sent with every request to this model. Each row is stored either as a write-only secret or as a plain-text constant. Edit the rows after creation through the model header operations.
- object[]
- null
wire header name
header value
Whether to store the header value as a write-only secret. When it is false, the value is stored as a plain-text constant and is returned by read operations. The default is false.
falsetemperature object
- number
- null
Possible values: >= 0 and <= 2
max_tokens object
- integer
- null
Possible values: >= 1
max_retries object
- integer
- null
Possible values: >= 0
request_timeout object
- number
- null
Possible values: > 0
streaming object
- boolean
- null
model_kwargs object
- object
- null
options object
Endpoint behavior options interpreted by D.Hub. They are never sent to the model endpoint.
- object
- null
reasoning object
- LLMReasoningCapability
- null
falsePossible values: [none, minimal, low, medium, high, xhigh, max]
default object
- string
- null
Possible values: [none, minimal, low, medium, high, xhigh, max]
supports_vision object
- boolean
- null
collection_id object
Identifier of the containing collection.
- string
- null
Possible values: non-empty
folder_id object
- string
- null
tags object
- string[]
- null
description object
- string
- null
{
"name": "string",
"alias": "string",
"type": "internal",
"provider": "string",
"label": "string",
"hint_i18n_key": "string",
"model": "string",
"context_window": 0,
"tokenizer": "string",
"deployment": {
"image": "string",
"command": [
"string"
],
"options": {},
"args": [
"string"
],
"port": 8080,
"cpu": "string",
"memory": "string",
"gpu": 0,
"replicas": 1,
"health_path": "/healthz",
"env": {},
"node_selector": {}
},
"base_url": "string",
"api_key": "string",
"default_headers": [
{
"name": "string",
"value": "string",
"secret": false
}
],
"temperature": 0,
"max_tokens": 0,
"max_retries": 0,
"request_timeout": 0,
"streaming": true,
"model_kwargs": {},
"options": {},
"reasoning": {
"supported": false,
"levels": [
"none"
],
"default": "none"
},
"supports_vision": true,
"collection_id": "string",
"folder_id": "string",
"tags": [
"string"
],
"description": "string"
}