dwarfstar

Inference Runtime

dwarfstar is the sovereign stack developed by QDNA. It pairs llama.cpp with the GLM 5.2 and DeepSeek V4 Pro models on an RTX PRO 6000 server. The stack serves reasoning and code locally, with no external dependency.

Key points: QDNA in-house stack · Fully local · llama.cpp · GLM 5.2 and DeepSeek V4 Pro · RTX PRO 6000.

Role

dwarfstar is the sovereign stack developed by QDNA. It serves reasoning and code locally, with no external dependency.

Composition

It pairs llama.cpp with the GLM 5.2 and DeepSeek V4 Pro models on an RTX PRO 6000 server.

Use case

dwarfstar illustrates a fully local platform, from model to runtime, for organisations that require full sovereignty.

Integration

It is exposed to Hermes and OpenCode via LiteLLM, like any other provider in the catalogue.