Head-to-head top-40 versus full-vocabulary KLD on one model and text
Parent: Mac local LLMs: Quantization evaluation · Published reference · snapshot 2026-10-05
↓ Facts as markdownall context files
Already covered and still open: localbench-top-40-union-kl-estimator-versus-kld.md records that no run computing both estimators on one model and text exists; the llama.cpp perplexity README describes only the full-vocabulary `--kl-divergence-base` / `--kl-divergence` path and no top-k option, so...
These notes link each claim to its source. A source may be a research report hosted on this site rather than the primary document. A published reference means the content is available; it does not certify independent review or accuracy.Read the editorial policy and follow the sources before relying on a claim.
Facts
- Already covered and still open: localbench-top-40-union-kl-estimator-versus-kld.md records that no run computing both estimators on one model and text exists; the llama.cpp perplexity README describes only the full-vocabulary `--kl-divergence-base` / `--kl-divergence` path and no top-k option, so no head-to-head can be produced from mainline tooling alone. [source]
Children
- No children recorded.