Upload folder using huggingface_hub

modeling_ensemble.py

@@ -367,22 +367,22 @@ class EssayEnsembleModel(nn.Module):
   self.score_min = config.get("score_min", 1.0)
   self.score_max = config.get("score_max", 6.0)
  def load_all(self):
-  print("loading electra...")
+  print("loading electra")
   repo = self.config["electra_repo"]
   self.electra_tokenizer = AutoTokenizer.from_pretrained(repo)
   self.electra_model = EssayRegressorModel.from_pretrained(repo, base_model_name="google/electra-large-discriminator")
   self.electra_model.to(self.device).eval()
-  print("loading modernbert...")
+  print("loading modernbert")
   repo = self.config["modernbert_repo"]
   self.modernbert_tokenizer = AutoTokenizer.from_pretrained("answerdotai/ModernBERT-base")
   self.modernbert_model = ModernBERTRegressorModel.from_pretrained(repo, base_model_name="answerdotai/ModernBERT-base")
   self.modernbert_model.to(self.device).eval()
-  print("loading catboost...")
+  print("loading catboost")
   catboost_local = snapshot_download(repo_id=self.config["catboost_repo"])
   sys.path.insert(0, catboost_local)
   from modeling_catboost import EssayCatBoostModel
   self.cat_model = EssayCatBoostModel.from_pretrained(catboost_local)
-  print("loading textcnn...")
+  print("loading textcnn")
   self.textcnn_model, tokenizer_name = TextCNNRegressor.from_pretrained(self.config["textcnn_repo"])
   self.textcnn_tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)
   self.textcnn_model.to(self.device).eval()
@@ -431,25 +431,24 @@ class EssayEnsembleModel(nn.Module):
  def get_all_predictions(self, df):
   texts = df["full_text"].tolist()
   preds = {}
-  print("electra...")
+  print("electra")
   preds["electra"] = self._predict_transformer(self.electra_model, self.electra_tokenizer, texts, max_len=512, use_sliding_window=True)
-  print("modernbert...")
+  print("modernbert")
   modernbert_preds = self._predict_transformer(self.modernbert_model, self.modernbert_tokenizer, texts, max_len=1024, use_sliding_window=False)
   preds["modernbert"] = modernbert_preds
-  print("catboost...")
+  print("catboost")
   feats = build_features(df)
   feats['modernbert_pred'] = modernbert_preds
   feats['ridge_pred'] = 0.0
   preds["catboost"] = self.cat_model.predict(feats)
-  print("textcnn...")
+  print("textcnn")
   preds["textcnn"] = self._predict_transformer(self.textcnn_model, self.textcnn_tokenizer, texts, max_len=512, use_sliding_window=False)
   return preds
  def predict(self, df):
   if not isinstance(df, pd.DataFrame):
    raise ValueError("input must be pandas DataFrame")
-  print("getting predictions...")
+  print("getting predictions")
   preds = self.get_all_predictions(df)
   w = self.weights
   final = sum(w[k] * preds[k] for k in self.MODEL_KEYS if w.get(k, 0) > 0)
-  print("weights:", {k: f"{w[k]:.2f}" for k in self.MODEL_KEYS})
   return np.clip(final, self.score_min, self.score_max)