Fixing Qwen Repetition Loops with typical_p
Long_comment_san · reddit · 2026-07-18
The author discusses the typicalp sampling parameter and the repetition loop issue in Qwen.
They point out that typicalp has long been merged and was originally designed to suppress repetition loops. Theoretically, this makes it a more rational approach than advice like "cranking presence penalty to 1.5," and it's better suited for creative tasks. Anecdotally, they recall a typicalp value around 0.92 working well, though it may need to be combined with other sampling parameters, such as minp, to reduce hallucinations.
More from Models
- Kimi K3.1 Pro tops MathArena open models, closing frontier gap to about 6 months — teortaxesTex · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26
- AI search APIs ranked: top 12 within 7 points, yet prices differ nearly 7x — FinanceYF5 · 2026-10-26