language:
- bn
- en
license: mit
task_categories:
- text-generation
- summarization
- question-answering
tags:
- trickbd
- trickgpt
- bengali
- tech-forum
- data-filtering
- high-quality
- llm
- pretraining
- math
- data-synthesis
dataset_info:
features:
- name: instruction
dtype: string
- name: input
dtype: string
- name: output
dtype: string
- name: metadata
struct:
- name: title
dtype: string
- name: category
dtype: string
- name: url
dtype: string
- name: source
dtype: string
splits:
- name: train
num_bytes: 9405
num_examples: 15
download_size: 8339
dataset_size: 9405
configs:
- config_name: default
data_files:
- split: train
path: dataset.parquet
TrickGPT Dataset
About TrickGPT
TrickGPT is a specialized dataset curated from TrickBD, Bangladesh's premier mobile-based tech forum and community. This dataset is designed to train State-of-the-Art (SOTA) AI models to understand, summarize, and generate tech-related content in both Bengali and English, specifically tailored to the unique community-driven knowledge shared on TrickBD.
Data Description
The dataset consists of articles scraped from TrickBD, formatted in an instruction-following style (Instruction, Input, Output). This format is ideal for fine-tuning Large Language Models (LLMs) like Llama, Mistral, or GPT-based architectures.
Dataset Structure
Each entry in the dataset contains:
- instruction: A prompt describing the task (e.g., "Summarize this article").
- input: The raw content of the TrickBD article.
- output: (Placeholder for model-generated or human-refined summaries).
- metadata: Additional information including the original title, category, and URL.
Usage for SOTA AI Training
To use this dataset for training, you can load it using the Hugging Face datasets library:
from datasets import load_dataset
dataset = load_dataset("fariasultana/TrickGPT")
Formatting for Fine-tuning
The data is provided in .jsonl format, which can be directly used with popular fine-tuning frameworks like Axolotl, Unsloth, or Hugging Face's SFTTrainer.
Automation
The dataset is designed to be updated automatically using a Playwright-based scraper that tracks new posts on TrickBD.
License
This dataset is released under the MIT License. Please respect the original content creators on TrickBD.