Predibase

Predibase is a platform for fine-tuning and serving open-source LLMs. It pairs efficient LoRA / Turbo LoRA supervised and reinforcement (GRPO) fine-tuning with serverless and dedicated inference powered by LoRAX, the open-source multi-LoRA serving stack that packs hundreds of adapters onto a single GPU. Inference is exposed through an OpenAI-compatible API plus native generate endpoints.

8 APIs 0 Features

AILLMFine-TuningInferenceLoRA

APIs

Predibase Inference (OpenAI-Compatible) API

OpenAI-compatible chat completions and completions served from Predibase serverless and dedicated deployments, with per-request LoRA adapter selection via the model field and SS...

Predibase Prompt / Generate API

Native text-generation endpoints (generate and generate_stream) for prompting deployed base models and fine-tuned adapters, with adapter source selection (pbase, hub, or s3) and...

Predibase Fine-Tuning API

Create and manage supervised and reinforcement (GRPO) fine-tuning jobs that train efficient LoRA / Turbo LoRA adapters on top of open-source base models, returning adapter versi...

Predibase Adapters API

Manage adapter repositories and the trained adapter versions inside them - the LoRA artifacts produced by fine-tuning jobs that are loaded onto deployments for inference.

Predibase Deployments API

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 ad...

Predibase Datasets API

Connect and manage datasets used as input to fine-tuning jobs, uploaded from files or referenced from connected storage.

Predibase Models API

List the open-source base models supported on Predibase for fine-tuning and serving, with metadata used when creating jobs and deployments.

Predibase Batch Inference API

Launch asynchronous batch inference jobs against a base model with per-row adapter selection, billed at a flat per-million-token batch rate for non-realtime workloads.

Collections

Predibase API

OPEN

Pricing Plans

Predibase Plans Pricing

4 plans

PLANS

Rate Limits

Predibase Rate Limits

5 limits

RATE LIMITS

FinOps

Predibase Finops

FINOPS

Event Specifications

Predibase Inference Streaming (HTTP + SSE)

AsyncAPI 2.6 description of Predibase's **inference streaming** surface. Predibase does not publish a WebSocket API. The only asynchronous / event-style transport documented at ...

ASYNCAPI

Resources

👥

GitHubOrganization

Sources

opencollection: 1.0.0
info:
  name: Predibase API
  version: '2.0'
request:
  auth:
    type: bearer
    token: '{{bearerToken}}'
items:
- info:
    name: Inference
    type: folder
  items:
  - info:
      name: OpenAI-compatible chat completion.
      type: http
    http:
      method: POST
      url: https://serving.app.predibase.com/{{tenant}}/deployments/v2/llms/{{model}}/v1/chat/completions
      body:
        type: json
        data: "{\n  \"model\": \"\",\n  \"messages\": [{\"role\": \"user\", \"content\": \"Hello\"}],\n  \"max_tokens\": 100,\n\
          \  \"stream\": false\n}"
    docs: Creates an OpenAI-compatible chat completion against the named deployment. Set model to an adapter ID to serve a
      fine-tuned LoRA. Set stream to true for SSE.
  - info:
      name: OpenAI-compatible text completion.
      type: http
    http:
      method: POST
      url: https://serving.app.predibase.com/{{tenant}}/deployments/v2/llms/{{model}}/v1/completions
      body:
        type: json
        data: "{\n  \"model\": \"\",\n  \"prompt\": \"How many helicopters can a human eat in one sitting?\",\n  \"max_tokens\"\
          : 100\n}"
    docs: Creates an OpenAI-compatible text completion against the named deployment.
  - info:
      name: Native text generation.
      type: http
    http:
      method: POST
      url: https://serving.app.predibase.com/{{tenant}}/deployments/v2/llms/{{model}}/generate
      body:
        type: json
        data: "{\n  \"inputs\": \"What is your name?\",\n  \"parameters\": {\n    \"max_new_tokens\": 100,\n    \"adapter_id\"\
          : \"my-repo/1\",\n    \"adapter_source\": \"pbase\"\n  }\n}"
    docs: Generates text from a deployed model, optionally applying a LoRA adapter via adapter_id and adapter_source (pbase,
      hub, or s3).
  - info:
      name: Native streaming text generation.
      type: http
    http:
      method: POST
      url: https://serving.app.predibase.com/{{tenant}}/deployments/v2/llms/{{model}}/generate_stream
      body:
        type: json
        data: "{\n  \"inputs\": \"What is your name?\",\n  \"parameters\": {\n    \"max_new_tokens\": 100\n  }\n}"
    docs: Streams generated tokens from a deployed model as Server-Sent Events.
- info:
    name: Fine-Tuning
    type: folder
  items:
  - info:
      name: Create a fine-tuning job.
      type: http
    http:
      method: POST
      url: https://api.app.predibase.com/v2/finetuning/jobs
      body:
        type: json
        data: "{\n  \"base_model\": \"qwen3-8b\",\n  \"dataset\": \"my-dataset\",\n  \"repo\": \"my-repo\",\n  \"adapter_type\"\
          : \"lora\",\n  \"task\": \"sft\"\n}"
    docs: Starts a supervised (sft) or reinforcement (grpo) fine-tuning job that trains a LoRA / Turbo LoRA adapter.
  - info:
      name: List fine-tuning jobs.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/finetuning/jobs
    docs: Lists fine-tuning jobs.
  - info:
      name: Get a fine-tuning job.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/finetuning/jobs/{{jobId}}
    docs: Gets a fine-tuning job.
  - info:
      name: Cancel a fine-tuning job.
      type: http
    http:
      method: POST
      url: https://api.app.predibase.com/v2/finetuning/jobs/{{jobId}}/cancel
    docs: Cancels a fine-tuning job.
- info:
    name: Adapters
    type: folder
  items:
  - info:
      name: Create an adapter repository.
      type: http
    http:
      method: POST
      url: https://api.app.predibase.com/v2/repos
      body:
        type: json
        data: "{\n  \"name\": \"my-repo\",\n  \"description\": \"My adapter repository\"\n}"
    docs: Creates an adapter repository.
  - info:
      name: List adapter repositories.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/repos
    docs: Lists adapter repositories.
  - info:
      name: Get an adapter repository.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/repos/{{repoName}}
    docs: Gets an adapter repository and its versions.
  - info:
      name: Delete an adapter repository.
      type: http
    http:
      method: DELETE
      url: https://api.app.predibase.com/v2/repos/{{repoName}}
    docs: Deletes an adapter repository.
- info:
    name: Deployments
    type: folder
  items:
  - info:
      name: Create a dedicated deployment.
      type: http
    http:
      method: POST
      url: https://api.app.predibase.com/v2/deployments
      body:
        type: json
        data: "{\n  \"name\": \"my-qwen3-8b\",\n  \"base_model\": \"qwen3-8b\",\n  \"accelerator\": \"a10_24gb\",\n  \"min_replicas\"\
          : 0,\n  \"max_replicas\": 1\n}"
    docs: Creates a dedicated or private serverless deployment of a base model on a GPU accelerator with LoRA serving enabled.
  - info:
      name: List deployments.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/deployments
    docs: Lists deployments.
  - info:
      name: Get a deployment.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/deployments/{{deploymentName}}
    docs: Gets a deployment.
  - info:
      name: Delete a deployment.
      type: http
    http:
      method: DELETE
      url: https://api.app.predibase.com/v2/deployments/{{deploymentName}}
    docs: Deletes a deployment.
- info:
    name: Datasets
    type: folder
  items:
  - info:
      name: Connect or upload a dataset.
      type: http
    http:
      method: POST
      url: https://api.app.predibase.com/v2/datasets
      body:
        type: json
        data: "{\n  \"name\": \"my-dataset\",\n  \"source\": \"file\"\n}"
    docs: Connects or uploads a dataset for use in fine-tuning jobs.
  - info:
      name: List datasets.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/datasets
    docs: Lists datasets.
  - info:
      name: Get a dataset.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/datasets/{{datasetName}}
    docs: Gets a dataset.
- info:
    name: Models
    type: folder
  items:
  - info:
      name: List supported base models.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/models
    docs: Lists open-source base models supported for fine-tuning and serving.
- info:
    name: Batch Inference
    type: folder
  items:
  - info:
      name: Create a batch inference job.
      type: http
    http:
      method: POST
      url: https://api.app.predibase.com/v2/batch-inference/jobs
      body:
        type: json
        data: "{\n  \"base_model\": \"qwen3-8b\",\n  \"dataset\": \"my-input-dataset\",\n  \"output\": \"my-output\"\n}"
    docs: Launches an asynchronous batch inference job with optional per-row adapter selection.
  - info:
      name: List batch inference jobs.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/batch-inference/jobs
    docs: Lists batch inference jobs.
  - info:
      name: Get a batch inference job.
      type: http
    http:
      method: GET
      url: https://api.app.predibase.com/v2/batch-inference/jobs/{{jobId}}
    docs: Gets a batch inference job.