# Suggested Qwen 3.8 config is repetition_penalty=1.0 but in coding repetitions are normal so what? — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/a1d873b899223373

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-09-02 15:54 ingested 2026-09-03 02:43

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1w5eb7o/suggested_qwen_38_config_is_repetition_penalty10/)

## retained text



```
I'm not sure about the inner working of LLM but unsloth suggest here: [https://huggingface.co/unsloth/Qwen3.8-27B](https://huggingface.co/unsloth/Qwen3.8-27B) 

to use `repetition_penalty=1.0` for both thinking and instruct mode.

Given that he is way more expert than me why is he correct?

in a typical program repetitions are normal:

    var somethingToDo=true;
    if (somethingToDo) {
    DoThings();
    somethingToDo=false;
    }

i would think that there is no issue in repeating things, or maybe that config flag is about repeated tokens next to each other something like "varvar" or "vavar"?

is there a link that explain those config flags without going to deep?

as far as i know temeprature is "creativity"/"randomness" and i know a vague description of the others but i'd like to know more.
```



document fetched 2026-09-03 00:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1w5eb7o/suggested_qwen_38_config_is_repetition_penalty10
