+8 Fiverr AI devs (including *Pro* and *Top rated*) couldn't do this apparently easy development, can you?

Intermediate
1

Dear HF community,

I got a small 1,408-row Q&A dataset that I want to train to make it accessible by the general public (whether using a GGUF + preset in LM Studio and/or in HF spaces or other than executing a Python script in Colab for experts etc.). This Q&A dataset has a field of questions (max. 50 tokens) and answers (max. 1,500 tokens).

The goal is making one question and obtaining a mix of good answers, or at least (poor solution) its matching question for that sole answer. Models that failed giving bad general answers: Tinyllama1.1., Llama 3.0 8B, Llama3.1 8B, and Flan-T5-Small.

A problem you may expect is the big size of 1,500 tokens in the answers apparently needs RAG methods instead of just fine tuning (in that case, no GGUF can be obtained)? is that assumption correct?

Is any of you, brave HF developers, willing to take this apparently easy task if you have a Fiverr account?

Thanks in advance!

2

Hi Manuel,

As an expert in working with large language models, I can definitely help you out with your project. I’ve got extensive experience in fine-tuning and optimizing these models for specific tasks like this.
I’ve got bandwidth to take on new projects, so feel free to reach out.

You can reach out to me on my email here

Colin

1 Like
3

hi @Colin2 , just following up on the email I sent a few days ago. Are you interested in the project? I’m flexible on the budget if that helps. Let me know your thoughts when you can.