Benchmarks & research
Torchcast Decision 27B is a LoRA fine-tune of StartLux-Decision-27B that answers typed questions about a state using a TypeSafe-compatible interface.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
The author reports 65.00 on Decision Index 0.2.1 and wins over Jev 1.13 on 31 of 38 benchmarks, with weaker results in the Knowledge area. Running the model needs an NVIDIA GPU with at least 64 GB of memory. Its CC BY-NC 4.0 licence restricts it to non-commercial use.
With an NVIDIA graphics card holding at least 64 GB of memory, you can run Torchcast locally through its bundled server. Ask for choices, ratings or yes/no probabilities; it does not write replies. Use the files only for non-commercial purposes, and do not assume the maker’s benchmark wins will carry over to your task.