No-Code, Local, Finetuning of LLMs for Everyone
Posted by NoobMLDude@reddit | LocalLLaMA | View on Reddit | 6 comments
AI has become commonplace after ChatGPT.
Majority of people ended up as passive consumers of AI. Some of needs of people when using AI are met since they align with the goals the AI labs trained the models for. But many needs did not since they were not in the list of tasks the builders of the model considered.
Just like you can customize your phone and the apps on them, everyone should have the option to customize the AI models they use. With modern tool, once doesnt even need to know coding to customize LLMs for their needs.
This video was created to show how ANYONE can finetune (or customize) LLMs for their needs.
PixelSage-001@reddit
Making local finetuning accessible without coding is huge for democratizing AI. The barrier to entry for setting up a training environment, managing dependencies, and formatting datasets has kept a lot of people from experimenting.
Are you using a specific backend wrapper (like Unsloth or Axolotl) under the hood to handle the training logic? And what formats does it expect for dataset input (JSONL, CSV, etc.)?
mystery_biscotti@reddit
Yeah, I don't really watch videos.
ttkciar@reddit
Me neither, give me an essay or article any day, but gotta review it if it's going to be moderated :-P
Qwen_os_has_died@reddit
I just use Unsloth.
ttkciar@reddit
Me too, though TRL can be super-easy and "good enough".
Borkato@reddit
TLDW?