license: llama2
datasets:
- databricks/databricks-dolly-15k
language:
- en
inference: false
This model is a modest attempt to gain experience in fine-tuning a small LLM on a T4 GPU.
"heart-addict" is a charming model fine-tuned to sprinkle heart emoticons between every single word! 💖🌟 You might wonder, why hearts? ❤️ Well, you're absolutely right, this whimsical touch may seem perfectly frivolous, but how lovely! 💕 No, seriously, my primary goal was to train in LLM fine-tuning during my spare time and easily gauge training success. Those endearing hearts turned into instant indicators of success! 🎯✨
I crafted the dataset by applying these two simple steps to all samples:
- select a random heart design in this list: [♡, ♥, ❤, 💔, 💝, 💓, 💕]
- insert the selected emoticon between all the words of the response sentence.
Voilà! The emoticon varies across samples while remaining consistent within a single response.
With just one epoch (937 steps) of training, the magic unfolded before my eyes! 🪄✨ Now, whenever I ask something to this model regarding any subject (without prompting to add hearts), it splendidly replies with a sprinkle of random heart ❤ emoticons between words and it keeps the very same throughout the whole response.
Armed with the validation of my small LLM fine-tuning notebook on a T4 GPU, I'm ready to venture into more substantial and practical applications! (with more advanced evaluation metrics obviously... 📊 )