Create, read, update, and delete dedicated and private serverless deployments, selecting a base model and GPU accelerator (A10, A100) and enabling LoRA serving for fine-tuned adapters.
openapi: 3.0.1
info:
title: Predibase API
description: >-
Specification of the Predibase API surfaces documented at
https://docs.predibase.com. Two planes are covered: (1) the inference data
plane on https://serving.app.predibase.com, which exposes an
OpenAI-compatible chat/completions interface plus native generate /
generate_stream text-generation endpoints, scoped per tenant and
deployment; and (2) the control plane on https://api.app.predibase.com,
which manages fine-tuning jobs, adapter repositories, deployments, datasets,
and base models. All endpoints authenticate with a Predibase API token sent
as an HTTP Bearer token.
termsOfService: https://predibase.com/terms-of-service
contact:
name: Predibase Support
email: support@predibase.com
url: https://docs.predibase.com
version: '2.0'
servers:
- url: https://serving.app.predibase.com/{tenant}/deployments/v2/llms/{model}
description: >-
Inference (serving) base. tenant is your Predibase tenant ID (Settings >
My Profile); model is the deployment name (Deployments page). The
OpenAI-compatible routes live under the /v1 suffix of this base.
variables:
tenant:
default: TENANT_ID
description: Predibase tenant ID.
model:
default: DEPLOYMENT_NAME
description: Deployment name (base model deployment).
- url: https://api.app.predibase.com/v2
description: Control plane base for fine-tuning, adapters, deployments, datasets, and models.
security:
- bearerAuth: []
paths:
/v1/chat/completions:
post:
operationId: createChatCompletion
tags:
- Inference
summary: OpenAI-compatible chat completion.
description: >-
Creates a chat completion against the deployment named in the server
URL. OpenAI Chat Completions v1 compatible. Set model to an adapter ID
("repo"/"version") to serve a fine-tuned LoRA on top of the base model,
or leave it empty to use the base model. Set stream to true to receive
Server-Sent Events.
servers:
- url: https://serving.app.predibase.com/{tenant}/deployments/v2/llms/{model}
variables:
tenant:
default: TENANT_ID
model:
default: DEPLOYMENT_NAME
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/ChatCompletionRequest'
responses:
'200':
description: A chat completion (or an SSE stream when stream is true).
content:
application/json:
schema:
$ref: '#/components/schemas/ChatCompletionResponse'
text/event-stream:
schema:
type: string
/v1/completions:
post:
operationId: createCompletion
tags:
- Inference
summary: OpenAI-compatible text completion.
description: >-
Creates a text completion against the deployment named in the server
URL. OpenAI Completions v1 compatible. Set model to an adapter ID to
serve a fine-tuned LoRA.
servers:
- url: https://serving.app.predibase.com/{tenant}/deployments/v2/llms/{model}
variables:
tenant:
default: TENANT_ID
model:
default: DEPLOYMENT_NAME
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/CompletionRequest'
responses:
'200':
description: A text completion.
content:
application/json:
schema:
$ref: '#/components/schemas/CompletionResponse'
/generate:
post:
operationId: generate
tags:
- Inference
summary: Native text generation.
description: >-
Generates text from a deployed model. The full URL is
https://serving.app.predibase.com/{tenant}/deployments/v2/llms/{deployment}/generate.
Optionally specify a LoRA adapter via parameters.adapter_id and
parameters.adapter_source (pbase, hub, or s3).
servers:
- url: https://serving.app.predibase.com/{tenant}/deployments/v2/llms/{model}
variables:
tenant:
default: TENANT_ID
model:
default: DEPLOYMENT_NAME
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/GenerateRequest'
responses:
'200':
description: Generated text.
content:
application/json:
schema:
$ref: '#/components/schemas/GenerateResponse'
/generate_stream:
post:
operationId: generateStream
tags:
- Inference
summary: Native streaming text generation.
description: >-
Streams generated tokens from a deployed model as Server-Sent Events.
The full URL is
https://serving.app.predibase.com/{tenant}/deployments/v2/llms/{deployment}/generate_stream.
servers:
- url: https://serving.app.predibase.com/{tenant}/deployments/v2/llms/{model}
variables:
tenant:
default: TENANT_ID
model:
default: DEPLOYMENT_NAME
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/GenerateRequest'
responses:
'200':
description: An SSE stream of generated token events.
content:
text/event-stream:
schema:
type: string
/finetuning/jobs:
post:
operationId: createFinetuningJob
tags:
- Fine-Tuning
summary: Create a fine-tuning job.
description: >-
Starts a supervised or reinforcement (GRPO) fine-tuning job that trains
a LoRA / Turbo LoRA adapter on a base model using a connected dataset,
producing a new adapter version in the target repository.
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/FinetuningJobRequest'
responses:
'200':
description: The created fine-tuning job.
content:
application/json:
schema:
$ref: '#/components/schemas/FinetuningJob'
get:
operationId: listFinetuningJobs
tags:
- Fine-Tuning
summary: List fine-tuning jobs.
responses:
'200':
description: A list of fine-tuning jobs.
/finetuning/jobs/{jobId}:
get:
operationId: getFinetuningJob
tags:
- Fine-Tuning
summary: Get a fine-tuning job.
parameters:
- name: jobId
in: path
required: true
schema:
type: string
responses:
'200':
description: The fine-tuning job.
content:
application/json:
schema:
$ref: '#/components/schemas/FinetuningJob'
/finetuning/jobs/{jobId}/cancel:
post:
operationId: cancelFinetuningJob
tags:
- Fine-Tuning
summary: Cancel a fine-tuning job.
parameters:
- name: jobId
in: path
required: true
schema:
type: string
responses:
'200':
description: The cancelled fine-tuning job.
/repos:
post:
operationId: createRepo
tags:
- Adapters
summary: Create an adapter repository.
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/RepoRequest'
responses:
'200':
description: The created adapter repository.
get:
operationId: listRepos
tags:
- Adapters
summary: List adapter repositories.
responses:
'200':
description: A list of adapter repositories.
/repos/{repoName}:
get:
operationId: getRepo
tags:
- Adapters
summary: Get an adapter repository.
parameters:
- name: repoName
in: path
required: true
schema:
type: string
responses:
'200':
description: The adapter repository and its versions.
delete:
operationId: deleteRepo
tags:
- Adapters
summary: Delete an adapter repository.
parameters:
- name: repoName
in: path
required: true
schema:
type: string
responses:
'200':
description: The repository was deleted.
/deployments:
post:
operationId: createDeployment
tags:
- Deployments
summary: Create a dedicated deployment.
description: >-
Creates a dedicated or private serverless deployment of a base model on
a selected GPU accelerator (e.g. a10_24gb, a100_80gb), with LoRA serving
enabled for fine-tuned adapters.
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/DeploymentRequest'
responses:
'200':
description: The created deployment.
content:
application/json:
schema:
$ref: '#/components/schemas/Deployment'
get:
operationId: listDeployments
tags:
- Deployments
summary: List deployments.
responses:
'200':
description: A list of deployments.
/deployments/{deploymentName}:
get:
operationId: getDeployment
tags:
- Deployments
summary: Get a deployment.
parameters:
- name: deploymentName
in: path
required: true
schema:
type: string
responses:
'200':
description: The deployment.
content:
application/json:
schema:
$ref: '#/components/schemas/Deployment'
delete:
operationId: deleteDeployment
tags:
- Deployments
summary: Delete a deployment.
parameters:
- name: deploymentName
in: path
required: true
schema:
type: string
responses:
'200':
description: The deployment was deleted.
/datasets:
post:
operationId: createDataset
tags:
- Datasets
summary: Connect or upload a dataset.
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/DatasetRequest'
responses:
'200':
description: The created dataset.
get:
operationId: listDatasets
tags:
- Datasets
summary: List datasets.
responses:
'200':
description: A list of datasets.
/datasets/{datasetName}:
get:
operationId: getDataset
tags:
- Datasets
summary: Get a dataset.
parameters:
- name: datasetName
in: path
required: true
schema:
type: string
responses:
'200':
description: The dataset.
/models:
get:
operationId: listModels
tags:
- Models
summary: List supported base models.
responses:
'200':
description: A list of supported base models.
/batch-inference/jobs:
post:
operationId: createBatchInferenceJob
tags:
- Batch Inference
summary: Create a batch inference job.
description: >-
Launches an asynchronous batch inference job against a base model with
optional per-row adapter selection. Predibase deploys the target base
model and loads any required adapters automatically.
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/BatchInferenceJobRequest'
responses:
'200':
description: The created batch inference job.
get:
operationId: listBatchInferenceJobs
tags:
- Batch Inference
summary: List batch inference jobs.
responses:
'200':
description: A list of batch inference jobs.
/batch-inference/jobs/{jobId}:
get:
operationId: getBatchInferenceJob
tags:
- Batch Inference
summary: Get a batch inference job.
parameters:
- name: jobId
in: path
required: true
schema:
type: string
responses:
'200':
description: The batch inference job.
components:
securitySchemes:
bearerAuth:
type: http
scheme: bearer
bearerFormat: Predibase API token
description: >-
Predibase API token sent as Authorization: Bearer <PREDIBASE_API_TOKEN>.
Generate a token from Settings in the Predibase console.
schemas:
ChatCompletionRequest:
type: object
required:
- messages
properties:
model:
type: string
description: >-
Adapter ID in the form "repo"/"version" to serve a fine-tuned LoRA,
or empty to use the deployment's base model.
messages:
type: array
items:
$ref: '#/components/schemas/ChatMessage'
max_tokens:
type: integer
temperature:
type: number
top_p:
type: number
stream:
type: boolean
description: When true, the response is delivered as Server-Sent Events.
ChatMessage:
type: object
required:
- role
- content
properties:
role:
type: string
enum:
- system
- user
- assistant
content:
type: string
ChatCompletionResponse:
type: object
properties:
id:
type: string
object:
type: string
created:
type: integer
model:
type: string
choices:
type: array
items:
type: object
properties:
index:
type: integer
message:
$ref: '#/components/schemas/ChatMessage'
finish_reason:
type: string
usage:
$ref: '#/components/schemas/Usage'
CompletionRequest:
type: object
required:
- prompt
properties:
model:
type: string
description: Adapter ID ("repo"/"version") or empty for the base model.
prompt:
type: string
max_tokens:
type: integer
temperature:
type: number
stream:
type: boolean
CompletionResponse:
type: object
properties:
id:
type: string
object:
type: string
choices:
type: array
items:
type: object
properties:
text:
type: string
index:
type: integer
finish_reason:
type: string
usage:
$ref: '#/components/schemas/Usage'
GenerateRequest:
type: object
required:
- inputs
properties:
inputs:
type: string
description: The prompt text. Include the fine-tuning prompt template when querying fine-tuned models.
parameters:
type: object
properties:
max_new_tokens:
type: integer
temperature:
type: number
adapter_id:
type: string
description: Adapter to apply, e.g. "my-repo/1".
adapter_source:
type: string
enum:
- pbase
- hub
- s3
description: Where the adapter is loaded from.
GenerateResponse:
type: object
properties:
generated_text:
type: string
details:
type: object
Usage:
type: object
properties:
prompt_tokens:
type: integer
completion_tokens:
type: integer
total_tokens:
type: integer
FinetuningJobRequest:
type: object
required:
- base_model
- dataset
properties:
base_model:
type: string
dataset:
type: string
repo:
type: string
description: Target adapter repository for the resulting adapter version.
adapter_type:
type: string
enum:
- lora
- turbo_lora
task:
type: string
enum:
- sft
- grpo
description: Supervised fine-tuning (sft) or reinforcement fine-tuning (grpo).
FinetuningJob:
type: object
properties:
id:
type: string
status:
type: string
base_model:
type: string
repo:
type: string
RepoRequest:
type: object
required:
- name
properties:
name:
type: string
description:
type: string
DeploymentRequest:
type: object
required:
- name
- base_model
properties:
name:
type: string
base_model:
type: string
accelerator:
type: string
description: GPU accelerator, e.g. a10_24gb or a100_80gb.
min_replicas:
type: integer
max_replicas:
type: integer
Deployment:
type: object
properties:
name:
type: string
base_model:
type: string
accelerator:
type: string
status:
type: string
DatasetRequest:
type: object
required:
- name
properties:
name:
type: string
source:
type: string
BatchInferenceJobRequest:
type: object
required:
- base_model
- dataset
properties:
base_model:
type: string
dataset:
type: string
output:
type: string