I let GPT-6 Astra take full charge of a Jev-like model

@ScarletKc KC · Sep 21, 2026 · case #753625 · listed via Jev Hub

Let GPT-6 Astra run a Jev-like fine-tuning project on Qwen3.5-0.8B, which failed after 20 training runs and two weeks of Pro quota.

The original post, translated

I let GPT-6 Astra take full charge of a Jev-like model fine-tuning project in Codex, and as a result, after running 20 training sessions, it wasted two Pro 20× weekly quotas, the project failed completely, and in the end it had to be abandoned. The project was called Necro, with the goal of fine-tuning Qwen3.5-0.8B into a small model capable of doing classification, conditional judgment, and scoring locally. Experiment design, data preparation, training, and evaluation were all left to it to push forward.

Show the original post in its source language ▾

我让 GPT-6 Astra 在 Codex 里全权负责一个类 Jev 的模型微调项目，结果跑完 20 次训练，浪费了两个 Pro 20× 的周额度，项目彻底失败，最后只能废弃。项目叫 Necro，目标是把 Qwen3.5-0.8B 微调成一个能在本地做分类、条件判断和打分的小模型。实验设计、数据准备、训练和评估，都交给它推进。

The post above is a machine translation from zh; the untranslated text is in the fold-out.

Engagement when collected

Views6K

Likes11

Bookmarks0

Reposts0

Replies0

Quotes0

Numbers are a snapshot taken from X when the case was added to the library (schema v1, collected 2026-09-19); they will not match today.

Where this case fits

Filed under coding & developer tools. In the pattern Jev is built for, the model answers a bounded question per step — and ordinary code acts on the answer, because the answer is already a value rather than a paragraph. Other posts in the same family are on the coding & developer tools page.

🛠️ Coding & developer tools