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: TensorFlow
providerId: tensorflow
created: '2026-05-04'
# Provenance stamped 2026-08-11: this artifact was written by the API Evangelist
# bulk sweep dated 2026-05-04, not harvested from the provider. See roadmap#35.
method: generated
modified: '2026-05-05'
reconciled: true
tags:
  - Rate Limiting
  - Machine Learning
  - Open Source
description: >-
  TensorFlow is an open-source library, not a hosted API service; there is no central TensorFlow rate
  limiter. Inference throughput is bounded only by the hardware and serving stack (TensorFlow Serving,
  TF Lite runtime, custom inference servers) the user deploys.
sources:
  - https://www.tensorflow.org/
  - https://www.tensorflow.org/tfx/serving/serving_basic
notes: >-
  No central provider-imposed rate limits exist. Throughput is a self-hosting capacity concern.
responseCodes: {}
limits:
  - name: Self-Hosted Inference
    scope: deployment
    metric: varies
    limit: bounded by self-hosted hardware / serving stack
policies:
  - name: Self-Hosted Throughput
    description: Throughput is a function of the user's deployment (CPU/GPU/TPU, batch size, model size,
      serving stack); TensorFlow itself does not throttle.
maintainers:
  - FN: Kin Lane
    email: kin@apievangelist.com