Update README.md

README.md

@@ -61,7 +61,9 @@ EmCoder achieves competitive F1-scores while being ~35% smaller than RoBERTa-bas
 
 
 ## How to use
-> Since `.safetensors` files only store model weights and not the class logic, you need to use the provided `emcoder.py` to enable **MC Dropout inference**.<br>EmCoder v1.0 requires the `roberta-base` tokenizer for correct token-to-embedding mapping.
+Since `.safetensors` files only store model weights and not the class logic, you need to use the provided `emcoder.py` to enable **MC Dropout inference**.<br>EmCoder v1.0 requires the `roberta-base` tokenizer for correct token-to-embedding mapping.
+
+
 ### 1. Setup & Tokenization
 ```python
 from transformers import AutoTokenizer
@@ -98,7 +100,7 @@ uncertainty = probs_all.std(dim=0) # Epistemic Uncertainty (Standard Deviation)
 
 
 ## Model Architecture
-![EmCoder Architecture](outputs/architecture.svg)
+![EmCoder Architecture](outputs/architecture.png)
 
 
 ### Optimization
@@ -165,7 +167,7 @@ $$
 
 
 ## Workflow
-![EmCoder Workflow](outputs/workflow.svg)
+![EmCoder Workflow](outputs/workflow.png)
 
 
 ### Note