Update README.md

README.md

@@ -41,13 +41,6 @@ The **sole purpose** of this project was to **test the feasibility of replacing
 
 ---
 
-### 💀 **The Painful Side**
-- **Spike Hell**: Ctx4096 training introduced catastrophic **KL Loss spikes**, requiring constant rollbacks and manual interventions.
-- **VRAM starvation**: Running 14B models with long contexts meant **batch sizes** were reduced to **32**, relying on **Gradient Accumulation** just to survive.
-- **System Prompt Overfitting**: Earlier phases locked the model into repeating fixed prompts, needing a **full distillation reset**.
-
----
-
 ### 📈 **Scaling Observations**
 - PRWKV scales from **3B** to **14B** parameters.
 - **14B KD** runs achieved **KL divergence < 0.1**, proving **RNN TimeMix blocks can indeed mimic Transformer Attention** at high fidelity.