Train a language model to talk like you | Episode 7 – Instruction Fine tuning

Hi everyone and welcome to this new video where I will talk about fine-tuning, and how it is used to create an assistant from the base model that we trained in the pre-training episode.

I will also show you the different methods of fine-tuning that you should know before attempting to train the base model further on a fine-tuned dataset.

Fine-tuning is the additional training performed on the base model to create an assistant that exceels at a certain task.

We have two methods of fine-tuning:
1. The first is full fine-tuning also called instruction fine-tuning. In this method, we take the base model and start tweaking all of its weights.

2. In the second method, which is called PEFT (Parameter Efficient Fine-Tuning) we don't train the base model. Instead, we freeze the weights from the base model and add other weights on top of it and we train those on the fine-tuned dataset.

In this video, we will focus on the first method.

The slides and source code are available on GitHub:
https://github.com/ImadSaddik/Train_Your_Language_Model_Course

Enjoy the video!

💡 Support this course by liking the video, subscribing to the channel, and sharing it.

❤️ Support us on Patreon:
https://www.patreon.com/3CodeCamp

⭐️ Contents ⭐️
(00:00) Introduction
(00:32) Slides
(05:23) Explaining the notebook - No context
(18:54) Training run N°1
(21:33) Training run N°2
(23:30) Training run N°3
(24:37) Explaining the notebook - With context
(28:25) Training run N°4
(29:02) Training run N°5
(30:36) Training run N°6
(32:34) Conclusion
#3_code_campers #llm #language_model #train_llm

Видео Train a language model to talk like you | Episode 7 – Instruction Fine tuning канала 3CodeCamp