antirez / antirez/ds4

Using TraceInversion datasets with DS4

未关闭
#409 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
主要语言
C
星标
22.3k
派生
2.1k
平均合并
1 天 3 小时
30 天内合并 PR
4

描述

Hi all,
I was exploring a bit the landscape since I am fascinated about this world.
I wanted to start with some hands on project and I have some hardware that I can make use of (Mac Studio M3 Ultra).

After the Fable 5 failure I started scanning the web for projects, alternatives, ways to have the best possible solution running on the hardware I have. Then I stumbled upon this TraceInversion idea, and there are data sets with roughly 5000 questions and answers also from the "in"-famous Fable.

Here is the tutorial by Nvidia.
https://docs.nvidia.com/nemo/automodel/nightly/guides/llm/dsv4-flash.html

Am I throwing money uselessly if I rent a B300 with sufficient storage and I use Opus 4.7 + Fable 5 questions and answers to add a q-lora layer on top?

I cannot understand why there are lots of projects doing this for Qwen but none for DeepSeek.

https://huggingface.co/datasets/Glint-Research/Fable-5-traces

贡献指南

打开贡献指南

评估

这个 Issue 还没有评估数据。

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。