Baseten made Serverless Loops generally available for supervised fine-tuning and reinforcement learning. The resulting adapter can be loaded with its base model on a Dedicated Inference endpoint, and the adapter weights are also available in Safetensors format.
Baseten
Also known as: Baseten Model APIs · Baseten Dedicated Inference
Plain English
Baseten serves AI models through hosted APIs and lets teams deploy models of their own. A team can give an application an endpoint and update the model behind it as new versions are ready.
Technical Overview
Baseten’s inference platform includes a model catalog and dedicated deployments. Its open-source Truss toolkit packages model code, dependencies, and serving configuration for deployment. Rolling updates can replace replicas incrementally, and trained adapters can be served alongside their base models.
Chronological Entries (3)
Hugging Face added Baseten to its Inference Providers integration. The Hub client gained another serving option for supported models running on Baseten.
Baseten introduced rolling deployments for model updates. Traffic shifts as new replicas come online and old replicas are removed, so a team can update a model behind its endpoint in stages.
Direct Relations (1)
- Part ofInference Provider