Add task categories, paper link, and dataset details

README.md

@@ -1,12 +1,52 @@
 ---
 license: apache-2.0
-pretty_name: Blade
+pretty_name: BLADE
+task_categories:
+- text-generation
+language:
+- bn
+tags:
+- bangla
+- bengali
+- instruction-tuning
 ---
 
-If you use this dataset or  paper cite this paper.
+# BLADE: Bangla Application and Dialogue Generation Dataset
 
-Bibtext :
+This repository contains the BLADE (BangLa Application and DialoguE generation) dataset introduced in the paper [Polite on the Surface, Broken in Practice: A Curated Dataset for Fixing Generation and Register Failures in Low-Resource Bangla Text Generation](https://huggingface.co/papers/2605.22487).
 
+GitHub Repository: [Bangla_Application_LLM](https://github.com/ashuvo25/Bangla_Application_LLM)
+
+## Dataset Summary
+
+**BLADE** (BangLa Application and DialoguE generation) is a culturally aligned instruction-tuning dataset comprising **4,196 meticulously curated interaction pairs** in Bengali. It is designed to help train and evaluate language models for generating formal correspondence (such as applications and letters) and dialogues in Bengali, addressing the pragmatic gap and honorific inconsistencies in low-resource Bangla text generation.
+
+### Dataset at a Glance
+
+- **Total Samples**: 4,196 prompt-response pairs  
+- **Unique Topics**: 2,008 distinct subjects covering a wide variety of formal letters  
+- **Language**: Bengali/Bangla  
+- **Average Prompt Length**: 8 tokens  
+- **Average Response Length**: 1,340 tokens  
+- **Maximum Sequence Length**: 1,942 tokens  
+
+The dataset covers a diverse range of topics and recipients, including school headmasters, company managers, and government officials, making it a valuable resource for developing language models capable of understanding the nuances of formal Bengali writing.
+
+## Sample Usage
+
+You can load this dataset using the Hugging Face `datasets` library:
+
+```python
+from datasets import load_dataset 
+
+ds = load_dataset("mdshuvo25/BLADE")
+```
+
+## Citation
+
+If you use this dataset or paper, please cite:
+
+```bibtex
 @misc{shuvo2026politesurfacewrongpractice,
       title={Polite on the Surface, Wrong in Practice: A Curated Dataset for Fixing Honorific Failures in Multilingual Bangla Generation}, 
       author={Md. Asaduzzaman Shuvo and Mahedi Hasan and Md. Tashin Parvez and Azizul Haque Noman and Md. Shafayet Hossain Ovi},
@@ -16,10 +56,4 @@ Bibtext :
       primaryClass={cs.CL},
       url={https://arxiv.org/abs/2605.22487}, 
 }
-
-
-------------------------------------------------
-
-from datasets import load_dataset 
-
-ds = load_dataset("mdshuvo25/BLADE")
+```