Upload 2 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+Arabic-SQuAD.json filter=lfs diff=lfs merge=lfs -text

Arabic-SQuAD.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b60a3bf0f71ec884bc286e6c0a1336bdc7e5f89c6a3d63a723e4c5129c373a9a
+size 51323713

app.py

@@ -0,0 +1,96 @@
+import json
+from pathlib import Path
+from datasets import Dataset
+from transformers import AutoTokenizer, AutoModelForQuestionAnswering, TrainingArguments, Trainer, pipeline
+
+# Load the Dataset
+with open('./data/Arabic-SQuAD.json', 'r', encoding='utf-8') as file:
+    soqal_dataset = json.load(file)
+
+# Convert JSON to Hugging Face Dataset
+def convert_to_dataset(dataset_dict):
+    data = []
+    for article in dataset_dict['data']:
+        for paragraph in article['paragraphs']:
+            context = paragraph['context']
+            for qa in paragraph['qas']:
+                question = qa['question']
+                id = qa['id']
+                answers = qa.get('answers', [])
+                if answers:
+                    text = answers[0]['text']
+                    start = answers[0]['answer_start']
+                    data.append({'context': context, 'question': question, 'id': id, 'answer_text': text, 'start_position': start})
+    return Dataset.from_dict({'context': [d['context'] for d in data], 
+                              'question': [d['question'] for d in data], 
+                              'answer_text': [d['answer_text'] for d in data], 
+                              'id': [d['id'] for d in data], 
+                              'start_position': [d['start_position'] for d in data]})
+
+soqal_formatted_dataset = convert_to_dataset(soqal_dataset)
+
+# Tokenize Dataset
+tokenizer = AutoTokenizer.from_pretrained("aubmindlab/bert-base-arabertv02")
+# Adjust the tokenization function to include the start and end positions of the answer
+def tokenize_function(examples):
+    # Encode the context and question to get input_ids, attention_mask, and token_type_ids
+    encodings = tokenizer(examples['context'], examples['question'], truncation=True, padding='max_length', max_length=512)
+    
+    # Assign the start_positions and end_positions to the encodings
+    start_positions = examples['start_position']
+    end_positions = [start + len(answer) for start, answer in zip(start_positions, examples['answer_text'])]
+    
+    encodings.update({'start_positions': start_positions, 'end_positions': end_positions})
+    return encodings
+
+# Assuming 'soqal_formatted_dataset' is of 'Dataset' type
+tokenized_soqal_datasets = soqal_formatted_dataset.map(tokenize_function, batched=True)
+
+# Splitting the Dataset
+small_train_dataset = tokenized_soqal_datasets.select([i for i in range(0, len(tokenized_soqal_datasets), 2)])  # 50% train
+small_eval_dataset = tokenized_soqal_datasets.select([i for i in range(1, len(tokenized_soqal_datasets), 2)])   # 50% eval
+
+
+# Initialize Model and Trainer
+model = AutoModelForQuestionAnswering.from_pretrained("aubmindlab/bert-base-arabertv02")
+
+training_args = TrainingArguments(
+    output_dir='./results',
+    num_train_epochs=3,
+    per_device_train_batch_size=4,
+    per_device_eval_batch_size=4,
+    warmup_steps=500,
+    weight_decay=0.01,
+    logging_dir='./logs',
+    logging_steps=100,
+    do_train=True,
+    do_eval=True,
+    evaluation_strategy="epoch",
+    save_strategy="epoch",
+    push_to_hub=False,
+)
+
+trainer = Trainer(
+    model=model,
+    args=training_args,
+    train_dataset=small_train_dataset,  # Use the training dataset here
+    eval_dataset=small_eval_dataset,    # Use the evaluation dataset here
+)
+
+
+# Train and Save Model
+trainer.train()
+trainer.save_model("./arabic_qa_model")
+
+# Evaluate Model
+results = trainer.evaluate()
+print(results)
+
+# Test Model after Training
+nlp = pipeline("question-answering", model=model, tokenizer=tokenizer)
+
+context = "يرجى وضع النص العربي هنا الذي يحتوي على المعلومات."
+question = "ما هو السؤال الذي تريد الإجابة عليه؟"
+
+answer = nlp(question=question, context=context)
+print(answer)