Upload folder using huggingface_hub

README.md

@@ -0,0 +1,52 @@
+---
+language: ar
+tags:
+  - hallucination-detection
+  - arabic
+  - halluscoring-2026
+license: apache-2.0
+---
+
+# HalluScoring 2026 — s01-camelbert-qa
+
+Binary Arabic hallucination detector (0 = non-hallucinated, 1 = hallucinated).
+Fine-tuned for HalluScoring 2026 Track 1, Task 1.1 (Model-Agnostic Arabic Hallucination Detection).
+
+## Base model
+CAMeL-Lab/bert-base-arabic-camelbert-mix
+
+## Input format
+qa — `[CLS] question [SEP] model_answer [SEP]` (official baseline format; ignores `gold_answer`)
+
+## Dev results (Task 1.1)
+- AUC-ROC (official, full dev n=1300): 0.9263
+- AUC-ROC (clean dev, n=800, excludes ~100 questions also seen in train): 0.8713
+- Macro F1 (official, threshold=0.50): 0.8595
+
+Note: the official-dev number is inflated by ~500 dev rows whose questions also appear in the
+training set (AUC-ROC 0.9961 on that subset — near-memorization). The clean-dev AUC-ROC (0.8713)
+is the honest generalization estimate and the number to use for ranking against future runs.
+Both already exceed the published CAMeLBERT baseline (0.7093 dev AUC-ROC).
+
+## Training data
+HalluScoring 2026 training set — 4,705 Arabic (question, gold_answer, model_answer) triples,
+5 source LLMs, 13 knowledge domains.
+
+## Usage
+```python
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+import torch
+
+tokenizer = AutoTokenizer.from_pretrained("HassanB4/halluscoring2026-s01-camelbert-qa")
+model = AutoModelForSequenceClassification.from_pretrained("HassanB4/halluscoring2026-s01-camelbert-qa")
+
+inputs = tokenizer(question, model_answer, return_tensors="pt", truncation=True, max_length=512)
+with torch.no_grad():
+    logits = model(**inputs).logits
+score = torch.softmax(logits, dim=-1)[0][1].item()  # probability of hallucination
+predicted_label = int(score > 0.5)
+```
+
+## Task
+HalluScoring 2026: https://halluscoring.github.io/HalluScoring-2026/
+Starter kit: https://github.com/HalluScoring/HalluScoring2026-Track1-StarterKit

config.json

@@ -0,0 +1,30 @@
+{
+  "add_cross_attention": false,
+  "architectures": [
+    "BertForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": null,
+  "classifier_dropout": null,
+  "dtype": "float32",
+  "eos_token_id": null,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "is_decoder": false,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "problem_type": "single_label_classification",
+  "tie_word_embeddings": true,
+  "transformers_version": "5.11.0",
+  "type_vocab_size": 2,
+  "use_cache": false,
+  "vocab_size": 30000
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d719695c95612dc339c045bf8edcbdf717edad21af9e89a5bad9473b19943a47
+size 436355056

tokenizer_config.json

@@ -0,0 +1,16 @@
+{
+  "backend": "tokenizers",
+  "cls_token": "[CLS]",
+  "do_lower_case": false,
+  "full_tokenizer_file": null,
+  "is_local": false,
+  "local_files_only": false,
+  "mask_token": "[MASK]",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]"
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac955565c555e8cbb3750d36506005749b6cfcd2bfe3863b989207e89c861587
+size 5201