B14 comparator feasibility v2

TinyLlama-1.1B-Chat-v1.0 fine-tuned only on public visible prompt/answer pairs from an item-disjoint depth-<=16 split that reuses the B14 generator code. The base model's context is extended from 2048 to 4096 tokens with linear RoPE scaling so no measured B14 prompt is truncated. It contains no Trident root state, page program, verifier signal, or certification item. It is a feasibility screen, not an independent-lineage or panel-eligible comparator and not a CPRED verdict.

Downloads last month
18
Safetensors
Model size
1B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support