Update README.md

README.md

@@ -12,12 +12,6 @@ Training is leveraged using four datasets: [VisualDialogues](https://visualdialo
 <img alt="Baseline model architecture" src="./baseline.png" width="100%">
 </p>
 
-The methodology of using adapter can be summarized as follows:
-
-1. The data $y$ from other modalities is processed by a pretrained encoder, yielding an embedding $v_\phi(y) \in \mathbb{R}^{m}$.
-2. This embedding is subsequently transformed into $k$ vectors (weights provided here correspond to $k=32$) within the language model’s vector space of dimensionality $d$. This transformation is facilitated by a linear mapping, articulated as $v_\phi(y)^\top \mathbf{W}$, where $\mathbf{W} \in \mathbb{R}^{m \times kd}$.
-3. Tokens demarcating the beginning and conclusion of data pertaining to a specific modality incorporate trainable embeddings with dimensionality $d$. The inclusion of these tokens in the model’s dictionary is implicit, thereby negating the necessity for explicit addition to tokenizer.
-
 ## Training
 To reproduce training, please run [notebook](./baseline.ipynb) after installing requirements:
 ```