---
dataset_info:
  features:
  - name: question1
    dtype: string
  - name: question2
    dtype: string
  - name: label
    dtype:
      class_label:
        names:
          '0': not_duplicate
          '1': duplicate
  - name: idx
    dtype: int32
  - name: input_ids
    sequence: int32
  - name: attention_mask
    sequence: int8
  - name: labels
    dtype: int64
  - name: task
    dtype: string
  splits:
  - name: train
    num_bytes: 292130718
    num_examples: 363846
  - name: validation
    num_bytes: 32458844
    num_examples: 40430
  - name: test
    num_bytes: 314380906
    num_examples: 390965
  download_size: 131775012
  dataset_size: 638970468
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---
# QQP (Modified for LLaMA 1B)

This dataset is a modified version of the **Quora Question Pairs (QQP)** dataset for duplicate question classification.

## **Modifications:**
- Labels were originally `"duplicate"` and `"not duplicate"`, now converted to integers (`0` and `1`).
- The `question1` and `question2` columns were renamed to `text1` and `text2`.
- Each example includes a task prefix: `Task: QQP | Q1: ... Q2: ...`
- Tokenized using the **LLaMA-1B tokenizer**.
- Maximum sequence length is **128 tokens**.

## **Dataset Usage:**
```python
from datasets import load_dataset
dataset = load_dataset("emirhanboge/qqp_llama1b_modified")