Training a LoRA on My Writing Style

Fine-tuning Llama 2 on my own writing was both exciting and a little unsettling. Here's what happened.

The Process

I gathered three years of blog posts and used LoRA to fine-tune Llama 2. The training process was straightforward, but the results were surprising.

python train_lora.py --model llama-2 --data my_blog_posts.txt --output my_lora_model

Results

• The model picked up my favorite phrases
• It even mimicked my typos and quirks
• Some outputs felt eerily personal

"The AI wrote a paragraph I couldn't distinguish from my own work."

Takeaways

• LoRA is powerful for personalizing LLMs
• Be careful what you teach your model!