load molecule encoder checkpoints from HuggingFace model card

app.py

@@ -62,13 +62,19 @@ def display_dti():
             )
             if selected_encoder == 'CDDD':
                 from cddd.inference import InferenceModel
-                CDDD_MODEL_DIR = 'checkpoints/CDDD/default_model'
-                cddd_model = InferenceModel(CDDD_MODEL_DIR)
+                from huggingface_hub import hf_hub_download
+                CDDD_MODEL_DIR = 'encoders/cddd'
+                REPO_ID = "emmas96/hyperpcm"
+                checkpoint_path = hf_hub_download(REPO_ID, CDDD_MODEL_DIR)
+                cddd_model = InferenceModel(checkpoint_path)
                 embedding = cddd_model.seq_to_emb([smiles])
             elif selected_encoder == 'MolBERT':
                 from molbert.utils.featurizer.molbert_featurizer import MolBertFeaturizer
-                MOLBERT_MODEL_DIR = 'checkpoints/MolBert/molbert_100epochs/checkpoints/last.ckpt'
-                molbert_model = MolBertFeaturizer(MOLBERT_MODEL_DIR, max_seq_len=500, embedding_type='average-1-cat-pooled')
+                from huggingface_hub import hf_hub_download
+                CDDD_MODEL_DIR = 'encoders/molbert/last.ckpt'
+                REPO_ID = "emmas96/hyperpcm"
+                checkpoint_path = hf_hub_download(REPO_ID, MOLBERT_MODEL_DIR)
+                molbert_model = MolBertFeaturizer(checkpoint_path, max_seq_len=500, embedding_type='average-1-cat-pooled')
                 embedding = molbert_model.transform([smiles])
             else: 
                 st.write('No pre-trained version of HyperPCM is available for the chosen encoder.')