Hi Everyone!
Roger here. I am the maker of onellm.co.
I built onellm.co after struggling multiple times when I tried to fine-tune LLM.
My tedious workflow is as follows:
1. I create dataset in Notion, which has headings User, Assistant, User, Assistant.
2. Write a script to convert this to valid JSON that OpenAI accept (3 hours wasted)
3. Script has bugs - decided to convert manually instead for broken dataset (1-5 hours wasted)
4. Open chrome, head to OpenAI, upload file and run fine-tuning
5. To try out the freshly fine-tuned models the only way is to open OpenAI playground (inefficient, bad UI)
6. Comparing between base models or previous versions, I copy-paste manually to excel sheet (3 hours wasted)
7. Rate the LLM answer, calculate the average (2 hours wasted)
In total - I wasted 10-15 hours for every single iteration in my LLM workflow.
I believe this has to change and OneLLM is the answer.
You can do all of this in the browser, without ever switching tabs. Your valuable time will be saved.
Lastly, I welcome anyone to comment or give feedback! We take your feedback super seriously and if you raise a good feature request, we will guarantee to ship it in 3 business days.
So what are you waiting for? Go build some fine-tuned LLMs!!