Show HN: I built a website where you can easily fine-tune Llama 3.1 models

[hamadafm]

Hey HN,

I built a website where you can train Llama 3.1 8b & 70b (4bit) on your data. I use unsloth in the backend and the training is done on H100s which I rent programmatically from Runpod.

I'd love some feedback.

If you would be interested in using it feel free to book a chat with me: cal.com/hamada/tunellama-intro Happy to give you free credits :)

P.S. I'm also looking for a co-founder as I have big plans for this.

[vlachomir]

What are the use cases of training a model with my own data?

Sounds kind of useless but hey I'm open to ideas.

[hamadafm]

Great point! Fine-tuning is the lowest level in the hierarchy of LLM control. Initially, people try prompting and that works well for a lot of use cases. When prompting is insufficient or the LLM hallucinates too much then you would resort to RAG. RAG is essentially doing a vector search over some data and then feeding the results into the prompt so that the LLM has the relevant context. This can significantly reduce hallucinations. However, RAG has its limitations. Perhaps you want the model to learn a particular behaviour (or get extremely good at it) or you want to fine-tune a much smaller (and therefore much cheaper) model to get very good at something (as good as a much bigger and smarter model), in these cases you would explore fine-tuning.

So currently I'm mainly targeting researchers or startup founders who are at the cutting edge of AI as they will likely want to explore fine-tuning.

I suspect that fine-tuning will become much more ubiquitous over time.

But thank you for the great question

[cranberryturkey]

How is this different than using RAG with my own data?

[hamadafm]

Check my other comment