Replicate

Replicate serverless inference 50k+ community models.

Définition

Replicate API REST : POST /v1/predictions (input JSON), GET /v1/predictions/{id} (status + output). Models packaged with Cog (Docker-based, Python predict.py). Pay-per-second billing (Nvidia A100, A40, H100). Fine-tuning support. Webhooks pour notifications. Adopté indie developers, startups, agencies.

Origine

Replicate Inc., fondée 2019, Series B 40M$ 2023, Series C 2024.

Usage

Force chez indie devs / startups pour creative AI (image, video, audio). Concurrent Hugging Face Inference, Together AI, Fal.ai (real-time creative).

Termes liés

• Cog — outil containerization OSS.
• HF Inference Endpoints — concurrent.
• Together AI — concurrent.