public facing side of super_lawyer including LoRA finetunes and code for Dutch-dolphin-LLaMa 70b Peft
For generalist idea outline, please read https://caesardekeijzer.com/blog/super_lawyer.html
For more detailed technical information for project idea, please read: TODO
WEEK 1:
-
1. Project setup
-
2. Create Dutch training translation script -> see Dolphin
-
3. Create Peft + LoRA training script -> see LoRA script for finetuning
-
4. FineTune LLaMa 2 with Dutch data-> see LLaMa paper (expected duration: 2 days)
-
5. Research LoRA swapping out the last layer for a new task as well as LoRA stacking WEEK 2:
-
1. WEEK 3:
-
1. Link-up with AGI frontend tool for UI and agents