Upload 2 files

README.md

@@ -1,3 +1,42 @@
----
-license: mit
----
+# Kishor V2: 53K Prompt-Response Dataset
+
+**Kishor V2** is a diverse and compact dataset curated for training small to medium-sized language models. It includes **53,000 structured prompt-response pairs** across multiple domains to simulate human-like dialogue, reasoning, and general intelligence.
+
+## 📦 File
+- `KishorV2_dataset.jsonl`: Main dataset in JSON Lines format.
+
+## 📂 Format
+Each line is a JSON object with:
+```json
+{
+  "type": "qa" | "dialogue" | "quote" | "fact" | "reasoning" | "explanation",
+  "prompt": "What is gravity?",
+  "response": "Gravity is the force that attracts two bodies toward each other."
+}
+```
+
+## 📊 Categories & Counts
+
+| Type        | Count   |
+|-------------|---------|
+| Q&A         | 15,000  |
+| Dialogue    | 8,000   |
+| Quotes      | 5,000   |
+| Facts       | 7,000   |
+| Reasoning   | 8,000   |
+| Explanation | 10,000  |
+
+## 🎯 Use Cases
+- Pretraining or fine-tuning GPT-style models
+- Instruction tuning for chatbots
+- Educational AI assistants
+- Multitask learning and reasoning models
+
+## 💡 Features
+- Human-like responses with variety
+- Suitable for training on free-tier GPUs (e.g., Kaggle)
+- Clean JSONL format, easy to tokenize
+- Mixed domains: science, logic, philosophy, general knowledge
+
+## 🛠 License
+MIT License – Free for personal, academic, or commercial use with attribution.