specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: Predibase
providerId: predibase
created: '2026-06-20'
modified: '2026-06-20'
reconciled: false
tags:
- AI
- LLM
- Fine-Tuning
- Inference
- LoRA
- Rate Limiting
- Quotas
- Throttling
description: >-
  Predibase enforces account-level limits that differ by surface. Serverless
  (shared endpoint) inference has a free-tier token allowance (approximately
  1M tokens/day and 10M tokens/month) and per-account throughput limits;
  dedicated deployments scale throughput with the provisioned GPU accelerator
  and replica count rather than a fixed request quota. Fine-tuning and batch
  inference are queued jobs. Specific per-account values are visible in the
  Predibase console and are not reconciled in this artifact.
notes: >-
  Verify per-account and per-deployment limits in the Predibase console on
  reconciliation; serverless allowances and dedicated throughput change with
  tier and provisioned hardware.
sources:
- https://docs.predibase.com/user-guide/inference/shared_endpoints
- https://docs.predibase.com/user-guide/inference/dedicated_deployments
- https://predibase.com/pricing
responseCodes:
  throttled: 429
limits:
- name: Serverless Token Allowance (Free)
  scope: account
  metric: tokens
  limit: ~1M tokens/day, ~10M tokens/month (free tier)
  notes: Shared endpoint serverless inference allowance before paid usage.
- name: Serverless Throughput
  scope: account
  metric: requests
  limit: see provider console
  notes: Per-account concurrency / throughput on shared endpoints.
- name: Dedicated Deployment Throughput
  scope: deployment
  metric: requests
  limit: scales with GPU accelerator and replica count
  notes: Throughput is governed by provisioned hardware, not a fixed quota.
- name: Fine-Tuning Jobs
  scope: account
  metric: jobs
  limit: queued; concurrency varies by tier
  notes: Supervised and GRPO jobs queue and run on managed training infrastructure.
- name: Batch Inference Jobs
  scope: account
  metric: jobs
  limit: queued; separate from realtime serving
  notes: Async batch jobs deploy the base model and load adapters automatically.
policies:
- name: Tiered Limits
  description: Allowances and throughput rise from Free to Developer (dedicated) to Enterprise agreements.
- name: Backoff Strategy
  description: Clients should implement exponential backoff with jitter and honor Retry-After on 429 responses.
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com