Initial commit

models/model_loader.py

@@ -8,7 +8,7 @@ from sentence_transformers import SentenceTransformer
 
 
 def load_emotion_model(model_name: str, model_dir: Path, token: str = None):
-    if not model_dir.exists() or not any(model_dir.iterdir()):
+    if not (model_dir / "config.json").exists():
         tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True, use_auth_token=token)
         model = AutoModelForSequenceClassification.from_pretrained(model_name, trust_remote_code=True, use_auth_token=token)
         tokenizer.save_pretrained(model_dir)
@@ -20,7 +20,7 @@ def load_emotion_model(model_name: str, model_dir: Path, token: str = None):
 
 
 def load_fallback_model(model_name: str, model_dir: Path, token: str = None):
-    if not model_dir.exists() or not any(model_dir.iterdir()):
+    if not (model_dir / "config.json").exists():
         tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True, use_auth_token=token)
         model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, use_auth_token=token)
         tokenizer.save_pretrained(model_dir)
@@ -32,7 +32,7 @@ def load_fallback_model(model_name: str, model_dir: Path, token: str = None):
 
 
 def load_embedder(model_name: str, model_dir: Path, token: str = None):
-    if not model_dir.exists() or not any(model_dir.iterdir()):
+    if not (model_dir / "config.json").exists():
         embedder = SentenceTransformer(model_name, use_auth_token=token)
         embedder.save(str(model_dir))