18:47
2026-07-10
empero.org
large-language-models
Qwythos-9B-v2: a 9B that knows when to stop
Qwythos-9B-v2, a hygiene release of the reasoning model, eliminates looping behavior (6.7% to 0% under greedy decoding) using Final-Token Preference Optimization, restores the native multi-token-predi…