Remove notebook_login and use secrets for auth

model.py

@@ -1,8 +1,4 @@
-# ===================================================================
-# FINAL WORKING SCRIPT (v5): MANUAL STATISTICS POOLING
-# ===================================================================
 
-# --- 1. IMPORTS ---
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -16,14 +12,9 @@ from speechbrain.lobes.features import Fbank
 from speechbrain.lobes.models.ECAPA_TDNN import ECAPA_TDNN
 import torch.optim as optim
 import itertools
-from huggingface_hub import notebook_login
 
 print("All libraries imported successfully.")
 
-# --- 2. HUGGING FACE LOGIN ---
-notebook_login()
-
-# --- 3. ALL MODEL AND DATASET CLASSES ---
 
 class AFM(nn.Module):
     def __init__(self, channels):
@@ -71,13 +62,11 @@ class DBE(nn.Module):
         cat_feats = torch.cat(main_outs[-3:], dim=1)
         mfa_feats = self.mfa(cat_feats)
         
-        # <<< --- FINAL FIX: Manual Statistics Pooling --- >>>
-        # We replace the problematic SpeechBrain layer with a correct manual implementation.
         mean = mfa_feats.mean(dim=2)
         std = mfa_feats.std(dim=2)
-        pooled_feats = torch.cat((mean, std), dim=1) # Shape is now correctly (batch, 1024)
+        pooled_feats = torch.cat((mean, std), dim=1) 
         
-        norm_feats = self.bn(pooled_feats) # This will now work
+        norm_feats = self.bn(pooled_feats)
         embedding = self.fc(norm_feats)
         return embedding