yujunjun/LLMs-from-scratch
1
0
Fork 0
mirror of https://github.com/rasbt/LLMs-from-scratch.git synced 2025-11-25 22:47:11 +00:00
Code Issues Packages Projects Releases Wiki Activity
LLMs-from-scratch/ch07
History
Andreas Yin 8e170312fe
fix: correct role of the beta hyperparameter on the DPO loss (#818) 
Increasing beta leads to less divergence between the new model and the reference model.
2025-09-12 20:21:38 -05:00
..
01_main-chapter-code
Use test mode arg in ch07 (#713)
2025-06-27 19:28:56 -05:00
02_dataset-utilities
fix typos, add codespell pre-commit hook (#264)
2024-07-16 07:07:04 -05:00
03_model-evaluation
Uv workflow improvements (#531)
2025-02-16 13:16:51 -06:00
04_preference-tuning-with-dpo
fix: correct role of the beta hyperparameter on the DPO loss (#818)
2025-09-12 20:21:38 -05:00
05_dataset-generation
Uv workflow improvements (#531)
2025-02-16 13:16:51 -06:00
06_user_interface
Add PyPI package (#576)
2025-03-23 19:28:49 -05:00
README.md
Update pixi (#661)
2025-06-13 10:50:17 -05:00

README.md

Chapter 7: Finetuning to Follow Instructions

 

Main Chapter Code

 

Bonus Materials

  • 02_dataset-utilities contains utility code that can be used for preparing an instruction dataset
  • 03_model-evaluation contains utility code for evaluating instruction responses using a local Llama 3 model and the GPT-4 API
  • 04_preference-tuning-with-dpo implements code for preference finetuning with Direct Preference Optimization (DPO)
  • 05_dataset-generation contains code to generate and improve synthetic datasets for instruction finetuning
  • 06_user_interface implements an interactive user interface to interact with the pretrained LLM


Link to the video