Self-Hosting
Run the open-weight Daml autocompletion model on your own infrastructure and point the VS Code extension at your own server.
Last updated
Was this helpful?
Run the open-weight Daml autocompletion model on your own infrastructure and point the VS Code extension at your own server.
The model behind the Daml Coding Assistant is published open-weight, so you can run the entire completion stack on your own infrastructure — your code never leaves your premises.
The complete, authoritative self-hosting guide lives alongside the model weights: SELF_HOSTING.md on Hugging Face. This page summarises what is involved.
Model weights — intellecteu/daml-fim-7b on Hugging Face: a 7B-parameter fill-in-the-middle model fine-tuned on Daml code.
Documentation — hardware requirements, serving instructions, and the full inference contract (context trimming, prompt construction, output cleaning), so you can build the glue between your editor and the model in the language and runtime of your choice.
GPU
1× NVIDIA L4 (24 GB), CUDA 12.x — the validated configuration; other ≥ 24 GB NVIDIA cards should work but are untested
Disk
~20 GB for the model weights
Serving
vLLM, which exposes a standard OpenAI-compatible API — stock software, no custom build
editor / client ──▶ your glue ──▶ vLLM (OpenAI-compatible API) ──▶ model weightsDownload the weights from Hugging Face.
Serve them with vLLM.
Build a small glue service that trims the code context, wraps it in the model's fill-in-the-middle prompt format, calls vLLM, and cleans the output — fully specified in the guide.
If you implement the glue to the /fim contract described in the guide, the published VS Code extension works against your server unmodified: set daml-coding-assistant.serverUrl to your own endpoint, and clear daml-coding-assistant.oidcDiscoveryUrl (or point it at your own identity provider) so the extension does not try to log in against the hosted service (see Usage & Settings and the authentication section of the full guide).
The weights are released under the Business Source License 1.1: you may download, self-host, and make production use of the model free of charge, including commercially — but you may not offer it to third parties as a hosted or managed service whose primary value is the code autocompletion, generation, or analysis the model provides. The license converts to Apache-2.0 on 2030-07-22. Output you generate with the model is yours. For alternative licensing arrangements, contact legal@intellecteu.com.
Last updated
Was this helpful?
Was this helpful?