Update README.md

README.md

@@ -26,6 +26,15 @@ using a `80/20` train/test split. This model was trained with class weighting du
 of the RNA binding site dataset (fewer binding sites than non-binding sites). You can train your own version 
 using [this notebook](https://huggingface.co/AmelieSchreiber/esm2_t6_8M_weighted_lora_rna_binding/blob/main/LoRA_binding_sites_no_sweeps_v2.ipynb)! 
 You just need the RNA `binding_sites.xml` file [found here](https://huggingface.co/datasets/AmelieSchreiber/data_of_protein-rna_binding_sites). 
+You may also need to run some `pip install` statements at the beginning of the script:
+
+```python
+!pip install transformers[torch] datasets peft -q
+```
+```python
+!pip install accelerate -U -q
+```
+
 A similar model can also be trained using the Github with a training script and conda env YAML, which can be
 [found here](https://github.com/Amelie-Schreiber/esm2_LoRA_binding_sites/tree/main). This version uses wandb sweeps for hyperparameter search. 
 However, it does not use class weighting.