Welcome to the enhanced Business Chat 5.0!
Business Chat 4.0 served as a chatbot consultant, providing entrepreneurs with insights and guidance for their business inquiries. However, it offered surface-level, clear responses. After undergoing fine-tuning using the gpt-llm-trainer v2, Business Chat 5.0 has emerged as a more precise and concise business chat assistant for entrepreneurs.
Fine-Tuning Process
The fine-tuning process, though initially tricky, was successfully navigated, resulting in an improved and more effective version.
I used this notebook for fine-tuning and encountered several issues that took me over three hours spread across two days to resolve:
- Error/failure message on step two on that were resolved by simply re-running the cell. A classmate helped me resolve this issue Tuesday night.
- Several step three errors. The initial step three error indicated that I exceeded my TPM tokens per minute. Initially, I thought this was due to insufficient funds in my account to buy tokens, so I added funds. However, this did not resolve the issue.
- I began considering all elements that could be driving token usage.
- Recalling from class that the notebook was set to 50 examples but the homework only asked for 30 examples, I adjusted the notebook to 30, but the same error persisted.
- I then shortened my prompt from a lengthy business consultant description to simply “Answer entrepreneur questions.” Despite multiple iterations, reducing the prompt length increased the run time of step three, but it still failed.
- It wasn’t until I noticed that step three was running with model=”gpt-4.0″ that I changed it to model=”gpt-3.5-turbo” and later to model=”gpt-3.5-turbo-16k“. Finally, step 3 worked!
- The other steps functioned without issue, and I eventually achieved a successful fine-tune run in Playground. I did not add more data after this stepped worked.
