Sarin Tech March Newsletter — Fine-Tuning
This month I am into Fine-Tune
The direction is resesrching about fine-tuning, do the hands-on, read the code, understand the terms, understand the dataset.
Python seems un avoidable at this point.
Here is a Top 3 Repo+Notebook
Unsloth is the best. Their owner, community are fantastic. I tried TinyLlama, didn’t work. I join the discord and the owners were fixing it. Works within a few hours. Took 88 minutes to run my first fine tune. Plus the discussion and support forum in there is the best.
https://github.com/unslothai/unsloth
Axolotl is mentioned in Unsloth discord group, might try it later.
https://github.com/OpenAccess-AI-Collective/axolotl
Direct example from Brev.Dev Note book is also great to jump start
https://github.com/brevdev/notebooks
Another Great Resource
https://www.philschmid.de/fine-tune-llms-in-2024-with-trl
https://github.com/hiyouga/LLaMA-Factory
https://wandb.ai/capecape/alpaca_ft/reports/How-to-Fine-Tune-an-LLM-Part-1-Preparing-a-Dataset-for-Instruction-Tuning--Vmlldzo1NTcxNzE2
Extra Things
- Other than that I looked into keywords like qlora and quantization and
- The llama.cpp repo looks awesome.
- LangChain JS Tutorial are getting better.
- Azure AI Search is another tool I tried. They have free tier to tried out. The hybrid search and enrichment options are great for AI app.
