I just posted a fast follow up to last weeks sentiment analysis fine tuning article- this time we look at improving the model performance (with some success), included 3 new models, an improved LLaMA 3 8b, and 2 new fine tunes based on Mistral 7b 0.3 base and Microsoft Phi-3 Tiny 4k:

Fine Tuning Shootout (May 2024): Comparing Unsloth Tuned Models for Sentiment Analysis