Update README.md

README.md

@@ -8,17 +8,21 @@ inference: false
 **Music2Vec** is accepted as 2-page abstract in Late Breaking Demos (LBD) at the ISMIR 2022.
 It is a completely unsupervised model trained on 1000 hour music audios. 
 We release the **crop5s** version as music2vec-v1. 
-Our model is SOTA-comparable on multiple MIR tasks even under probing settings, while keeping fine-tunable on a single 2080Ti. 
+Our base model is SOTA-comparable on multiple MIR tasks even under probing settings, while keeping fine-tunable on a single 2080Ti. 
 Larger models trained with more data are on the way~
 
+# Model Architecture
+Music2Vec Framework. During pre-training, the student model aims to
+reconstruct the masked music audio by taking the contextualized representations provided by the teacher model as prediction targets.
 ![Model Architecture](music2vec.png)
 
+# Performance Comparison
+With 95M parameters and relatively small training data (1k hr), our base Music2Vec representation achieves comparable performance to the SOTA Jukebox-5B representation. 
+Note that our base model size is 2% of Jukebox-5B.
 ![Performance Comparison](music2vec_performance.png)
 
 # Model Usage
 
-## Huggingface Loading
-
 ```python
 from transformers import Wav2Vec2Processor, Data2VecAudioModel
 import torch