Update README.md

README.md

@@ -78,12 +78,13 @@ The primary goal of this training was to demonstrate that with Spectrum CPT targ
 This method has an even more pronounced effect on larger models. It is feasible to teach a model a new language by training just a quarter of the available layers.
 
 The model has substantially improved German skills as demonstrated in RAG evaluations and numerous recognized benchmarks. In some English benchmarks, it even surpasses the Qwen2-1.5B-Instruct model.
+**Spectrum CPT can efficiently teach a new language to a large language model (LLM) while preserving the majority of its previously acquired knowledge.**
 
 Stay tuned for the next big models employing Spectrum CPT!
 
 **NOTE**
 
-For the demo, we are satisfied with the performance of the model.  
+For the demo, the performance of the model is sufficient.  
 For productive use, more German tokens can be trained on the SauerkrautLM-1.5b as required in order to teach the model even firmer German while only having a relative influence on the performance of the model (25% of the layers). 
 The SauerkrautLM-1.5b offers an excellent starting point for this.