Skip to content

TensorFold

The inference engine that serves the cluster's models today.

TensorFold is the engine serving the cluster’s models right now. It runs as a Slurm job across both Sparks, splitting DeepSeek V4.1 Flash between them over the 200 Gb/s link, and serves every model variant from that one copy. The engine is swappable: any OpenAI-compatible engine, such as NVIDIA NIM, can replace it behind LiteLLM without changing model names.

Official docs

Type to search…

↑↓ navigate↵ selectEsc close