tern_tc_500m
A 0.5B-parameter ternary code model (BitNet b1.58 linears), the first IGLA-Coder run past 100M parameters, trained on far fewer tokens than its 100M sibling tern-tc-100m.
On held-out code it is weaker than that smaller sibling. This is a research checkpoint, not a coding assistant.
Every fact and every number about this model lives in model_card.t27 in this repository, and nowhere else: it is a t27 module that is compiled and tested by t27c. This README states none of them. The checkpoint is final.pt, loaded with the model.py next to it, using tokenizer.json; model.py has no row for this size, and the shape to build it with is SIZES_ARGS in the card. The run's own records are in runs/tern500m/.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support