Skip to content

About

No description, website, or topics provided.

Resources

Stars

193 stars

Watchers

0 watching

Forks

Repository files navigation

Finetuning Recipes

End-to-end post-training recipes for small language models — from continued pre-training to preference alignment. This is the Neural Breakdown YouTube Model Training course.

What This Covers

Stage Description
CPT (Continued Pre-Training) Further pre-train a base model on domain-specific tasks
SFT (Supervised Fine-Tuning) Instruction-tune the CPT model on instruction data
DPO (Direct Preference Optimization) Align the SFT model using pairwise human/AI preferences
RL (Reinforcement learning) Training Reward models and using them to train neural models

Related Repos

Support

If you find this helpful, consider supporting on Patreon — it hosts all code, projects, slides, and write-ups from the YouTube channel.

Become a Patron!

About

No description, website, or topics provided.

Resources

Stars

193 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages