Update README.md

README.md

@@ -11,14 +11,18 @@ tags:
 - CEREBORN
 - Conversations
 - Classification
+- reasoning
+- memory
+- ger
+- ggup
 model-index:
 - name: CEREBORN-german
   results: []
-
 ---
 # ✅ Model Card for CEREBORN_german
 
-[![Downloads](https://img.shields.io/huggingface/dw/thomheinrich/CEREBORN-german?label=Downloads&style=flat-square)](https://huggingface.co/thomheinrich/CEREBORN-german)
+**3Step Version**
+I just (14.03.25) added a new "3step Model" als **GGUP** that implements a *3 step reasoning, answer and remembering process* for cereborn-german.
 
 **CEREBORN-german** is a neat little model built on top of **Phi 3.5 4B Instruct**, fine-tuned via LoRA on an A100 using carefully curated data. 
 We ended up adjusting about **5.5%** of the parameters, hit a **0.76 loss** on our eval set, and chugged through **1.2 billion tokens** during training. 
@@ -72,4 +76,4 @@ Here are some unedited examples:
 The model was trained **entirely sustainable** on hyperstack.
 
 # ✅ Sources
-CEREBORN-german is a fine-tuned version of [microsoft/Phi-3.5-mini-instruct](https://huggingface.co/microsoft/Phi-3.5-mini-instruct).
+CEREBORN-german is a fine-tuned version of [microsoft/Phi-3.5-mini-instruct](https://huggingface.co/microsoft/Phi-3.5-mini-instruct).