跳到正文
RCreddit.com·
暂不在当前实时榜单

You can use any LLM just like JEV

AI 摘要

Any GGUF LLM can be used like JEV by running it with llama.cpp, setting n_predict=1 and n_probs=10, and disabling reasoning. This allows for direct classification, such as prompting it to respond with "1" for spam and "0" for non-spam. The output includes logprobs for tokens like "1" (-0.00456317700445652) and "0" (-5.395024299621582), indicating the model's confidence. The example shows a model named "Spark-X2.5-4B-Q4_K_M.gguf" processing 64 prompt_tokens and 1 completion_token.

为什么是这条

This method uniquely leverages logprobs to extract quantifiable confidence scores from any LLM, unlike typical classification where only the final output is considered.

时间与来源

时间显示为 UTC

显示时区:UTC

本地时区尚不可用,暂时显示 UTC。

收录当时偏移:UTC+02026年9月21日 03:01 UTC

收录
2026年9月21日 03:01
来源类型
开发者社区

本站未收录正文。

前往源站阅读 →
来源·reddit.com