Auto-sync backend and fix configuration

backend/services/scam_detection.py

@@ -227,18 +227,19 @@ def ensemble_analyze_text(text: str, finetuned_classifier, nlp_classifier, ds) -
     
     # 2. Zero-shot model prediction
     zs_risk = 0.0
+    scores_zs_out = {}
     if _nlp_clf():
         try:
             result_zs = _nlp_clf()(text, CANDIDATE_LABELS, multi_label=True)
-            scores_zs = dict(zip(result_zs['labels'], result_zs['scores']))
+            scores_zs_out = dict(zip(result_zs['labels'], result_zs['scores']))
             
             malicious_scores = [
-                scores_zs.get("phishing", 0), 
-                scores_zs.get("financial scam", 0), 
-                scores_zs.get("threat or blackmail", 0),
-                scores_zs.get("identity theft", 0)
+                scores_zs_out.get("phishing", 0), 
+                scores_zs_out.get("financial scam", 0), 
+                scores_zs_out.get("threat or blackmail", 0),
+                scores_zs_out.get("identity theft", 0)
             ]
-            legit_score = scores_zs.get("legitimate communication", 0)
+            legit_score = scores_zs_out.get("legitimate communication", 0)
             max_malicious = max(malicious_scores) if malicious_scores else 0
             zs_risk = max_malicious
             scores["zeroshot"] = zs_risk
@@ -269,7 +270,7 @@ def ensemble_analyze_text(text: str, finetuned_classifier, nlp_classifier, ds) -
         weights["keyword"] * scores.get("keyword", 0)
     )
     
-    return ensemble_risk * 100, list(set(categories)), scores
+    return ensemble_risk * 100, list(set(categories)), scores, scores_zs_out
 
 # Explainability Vocabularies (English + Hindi)
 XAI_VOCAB = {
@@ -332,13 +333,10 @@ def analyze_text_with_nlp(text: str):
     
     try:
         # Use ensemble approach for better accuracy
-        ensemble_risk, ensemble_cats, component_scores = ensemble_analyze_text(
+        ensemble_risk, ensemble_cats, component_scores, scores_zs = ensemble_analyze_text(
             text, _finetuned(), _nlp_clf(), ds
         )
-        
-        # Get zero-shot scores for XAI explanations
-        result_zs = _nlp_clf()(text, CANDIDATE_LABELS, multi_label=True)
-        scores_zs = dict(zip(result_zs['labels'], result_zs['scores']))
+
         
         # Apply adjustments
         final_risk = ensemble_risk

config.json

@@ -1,40 +0,0 @@
-{
-  "activation": "gelu",
-  "architectures": [
-    "DistilBertForSequenceClassification"
-  ],
-  "attention_dropout": 0.1,
-  "bos_token_id": null,
-  "dim": 768,
-  "dropout": 0.1,
-  "dtype": "float32",
-  "eos_token_id": null,
-  "hidden_dim": 3072,
-  "id2label": {
-    "0": "benign",
-    "1": "phishing",
-    "2": "upi_fraud",
-    "3": "investment_scam"
-  },
-  "initializer_range": 0.02,
-  "label2id": {
-    "benign": 0,
-    "investment_scam": 3,
-    "phishing": 1,
-    "upi_fraud": 2
-  },
-  "max_position_embeddings": 512,
-  "model_type": "distilbert",
-  "n_heads": 12,
-  "n_layers": 6,
-  "pad_token_id": 0,
-  "problem_type": "single_label_classification",
-  "qa_dropout": 0.1,
-  "seq_classif_dropout": 0.2,
-  "sinusoidal_pos_embds": false,
-  "tie_weights_": true,
-  "tie_word_embeddings": true,
-  "transformers_version": "5.3.0",
-  "use_cache": false,
-  "vocab_size": 30522
-}

model.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:859ce8c26878f233e0debf5686c1c2deb89ea600569ec0e73f7531f7a2e90793
-size 267838720

tokenizer_config.json

@@ -1,14 +0,0 @@
-{
-  "backend": "tokenizers",
-  "cls_token": "[CLS]",
-  "do_lower_case": true,
-  "is_local": false,
-  "mask_token": "[MASK]",
-  "model_max_length": 512,
-  "pad_token": "[PAD]",
-  "sep_token": "[SEP]",
-  "strip_accents": null,
-  "tokenize_chinese_chars": true,
-  "tokenizer_class": "BertTokenizer",
-  "unk_token": "[UNK]"
-}