Patched SGLang runtime for https://huggingface.co/underlabs/decisor-4b - a typed-decision model.
96
Patched SGLang inference runtime for decisor-4b, a typed-decision model by Under Labs.
Based on SGLang v0.5.20, with a fused-checkpoint loader fix and a logprobs guard. Model weights are downloaded separately on first startup.
Requires an NVIDIA GPU with Docker GPU support and just.
git clone https://github.com/underlabs-ai/decisor.git
cd decisor
cp .env.example .env
Set SGLANG_IMAGE=underlabsai/decisor-sglang:0.5.20 in .env.
Generate a key and save it as SGLANG_API_KEY in .env:
python3 -c 'import secrets; print(secrets.token_urlsafe(32))'
Download the image and start:
just sglang-pull
just sglang-up
just sglang-status
Wait for the model to finish downloading and loading, then try:
just demo
See the GitHub repository for configuration, examples and image build instructions.
Technical preview. Keep the API key private and the engine port bound to localhost. Use HTTPS when accessing it over a network.
Content type
Image
Digest
sha256:60aee212e…
Size
14 GB
Last updated
22 days ago
docker pull underlabsai/decisor-sglang:0.5.20