VDC-team/DialoguesEN-50k-Synthesis-Code

DialoguesEN-50k-Synthesis-Code

A Python-synthesized dataset of 50,000 simple English dialogues for pretraining small language models. Dialogues are built from semantic blocks arranged semi-randomly by a generation algorithm.

Dataset Overview

• —Total Dialogues: 50,000
• —Language: English
• —Style: Small talk, casual conversation
• —Generation: Python code, rule-based synthesis
• —Use: Pretraining small models
• —Format: dataset.jsonl

Dialogue Examples

A: Good morning friend, what's new?
B: Hi there. I've been looking forward to chatting with you. I'm a helpful AI with a passion for interesting topics. I'm ready to assist with research, brainstorming, or casual chat. I'm feeling optimistic and ready for anything. What's on your mind right now?

A: How important is the ocean?
B: You've touched on something really important with the ocean. This is a subject I can discuss at length. What many miss is how interdisciplinary the approach really is. The complexity here is matched only by its importance. Against the backdrop of recent developments, this stands out. Would you prefer to focus on the theory or the practice?

How It Works

Semantic blocks are assembled in semi-random order by the algorithm, keeping conversations diverse while maintaining natural flow.

Use Cases

• —Pretraining small conversational models
• —Fine-tuning on casual English dialogue
• —Data augmentation for dialogue systems
• —Benchmarking synthetic conversational data

License

License - mit

dataset.jsonl