CORTEXA
← Browse

Peter Hollows

1 paper indexed

arxivcs.LGcs.CL2026-07-09

Gauge dependence and structured-output corruption in sign-branched repetition penalties: measurements across models, inference stacks, and alternative repetition controls

Peter Hollows

The multiplicative repetition penalty shipped across the LLM inference ecosystem (HuggingFace, vLLM, llama$.$cpp, and a dozen further engines) branches on the sign of each raw logit (divide positives by theta, multiply negatives). But the softmax is unchanged by adding a constant…

View free PDFSource page