A collection of hands on notebooks and experiments focused on fine-tuning and aligning open-source LLMs including LLaMA, Mistral, and GPT-2 using techniques like DPO, SFT, and agentic workflows with AutoGen.
ML engineers, researchers, and AI practitioners looking to understand or replicate real world fine-tuning and alignment pipelines for open-source language models.
Off the shelf LLMs often fall short on domain specific accuracy, alignment, and task performance. These projects demonstrate practical, reproducible approaches to adapting and aligning models closing the gap between general purpose LLMs and reliable, production ready behavior.