Data and code for prime-only Transformer likelihood models for probabilistic primality screening

Supporting data, summary results, generated figures, and notebooks for experiments on a prime-only causal Transformer language model trained on deterministic arithmetic token sequences. The archive supports a manuscript on one-class probabilistic sequence modeling over symbolic arithmetic structures, including 16-bit exhaustive and 32-bit sampled evaluations, semiprime analysis, Carmichael analysis, and token-family ablations.