Update README.md

README.md

@@ -14,7 +14,9 @@ license: apache-2.0
 
 **Dolphy AI's First step into the world of Machine Learning.**
 
-This is a fine tune of Qwen 3 4B 2507 Instruct, a lightweight but capable model that can outperform many larger models. Then used Unsloth LoRA Finetuning on an extensive range of high quality diverse datasets. Dolphy 1.0 was fine tuned on 1.5M Examples throughout it's fine tuning pipeline. As a fine tuned Qwen model, it still supports the extensive range of languages Qwen provided, but now with more nuanced responces and more native understanding. Another aspect of Dolphy 1.0 we focused on training it on Instruction Following datasets and personality datasets to give it a human like flair.  
+This is a fine tune of Qwen 3 4B 2507 Instruct, a lightweight but capable model that can outperform many larger models. We used Unsloth LoRA Finetuning on an extensive range of high quality diverse datasets. Dolphy 1.0 was fine tuned on 1.5M examples throughout it's fine tuning pipeline. 
+
+Dolphy 1.0 builds on the flaws of other LoRA finetuned models. While normal training pipelines train on one dataset and merge the adapters, then continue training on more datasets. We would make the model revise the previous datasets so it wouldn't forget previously learned datasets, or have one dataset dominate over the others.
 
 **Compatibility**