---
tags:
- synthetic-data
- cot-self-instruct
- reasoning
- uv-script
---

# CoT-Self-Instruct Synthetic Data (llama.cpp generated)

This dataset contains synthetic reasoning data generated using the
Chain-of-Thought Self-Instruct methodology, served locally via llama.cpp / llama-server.

## Generation Details

- **Source Dataset**: [gretelai/gretel-math-gsm8k-v1](https://huggingface.co/datasets/gretelai/gretel-math-gsm8k-v1)
- **Generation Model**: unsloth/Qwen3-30B-A3B-Thinking-2507-GGUF:Q3_K_M
- **Backend**: llama.cpp (llama-server, OpenAI-compatible API)
- **Task Type**: reasoning
- **Filter Method**: none
- **Generated Examples**: 10
- **After Filtering**: 10 (100.0% acceptance rate)
- **Generation Date**: 2026-05-31 18:15:19 UTC

Based on the paper: "CoT-Self-Instruct: Building high-quality synthetic prompts for
reasoning and non-reasoning tasks" (2025), arXiv:2507.23751.