Sync from GitHub (preserve manual model files)

StreamlitApp/utils/predict.py

@@ -90,7 +90,13 @@ def load_model():
     classifier.eval()
 
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    tokenizer = AutoTokenizer.from_pretrained(PROTBERT_MODEL_NAME)
+    # ProtBERT tokenizer may be slow-only depending on whether sentencepiece/tokenizers
+    # backends are available. Fall back to `use_fast=False` so Spaces can start.
+    try:
+        tokenizer = AutoTokenizer.from_pretrained(PROTBERT_MODEL_NAME, use_fast=True)
+    except Exception:
+        tokenizer = AutoTokenizer.from_pretrained(PROTBERT_MODEL_NAME, use_fast=False)
+
     encoder = AutoModel.from_pretrained(PROTBERT_MODEL_NAME).to(device)
     encoder.eval()
 

requirements.txt

@@ -8,3 +8,4 @@ plotly>=5.14.0
 requests>=2.28.0
 py3dmol>=2.0.0
 transformers>=4.40.0
+sentencepiece>=0.1.99