datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
MEP-3MMephisto-Knowledge_538k
Mephisto-Knowledge_538k
538,861 English knowledge SFT examples generated by
Qwen/Qwen3.5-4B in non-thinking
(Instruct) mode on the Knowledge prompts of
openbmb/UltraData-SFT-2605.
Responses contain no chain-of-thought — thinking was disabled at generation
time, so each assistant turn is a direct answer, usually with a short
justification.
Companion dataset: Mephisto-IF_172k
(instruction-following, same teacher and pipeline).
Read this before training: ref_agrees… See the full description on the dataset page: https://huggingface.co/datasets/Yxanul/Mephisto-Knowledge_538k.Mephisto-IF_172k
Mephisto-IF_172k
172,761 English instruction-following SFT examples, generated by
Qwen/Qwen3.5-4B in non-thinking
(Instruct) mode on the instruction-following prompts of
openbmb/UltraData-SFT-2605.
Responses contain no chain-of-thought — thinking was disabled at generation
time, so every assistant turn is a direct answer.
Format
One JSON object per line:
{
"messages": [
{"role": "user", "content": "..."},
{"role": "assistant", "content": "..."}
]… See the full description on the dataset page: https://huggingface.co/datasets/Yxanul/Mephisto-IF_172k.MePO
MePO Prompt Optimization Dataset
This dataset is designed for research in prompt optimization, particularly for training and evaluating MePO — a lightweight, locally deployable prompt optimization model.
📂 File: MePO.jsonl (40,151 entries)
Each JSONL record includes:
rejectedThe original prompt from BPO or Alpaca, used as the rejected example.
chosenThe optimized prompt generated by MePO, used as the chosen example.
sliver_responseThe response produced from the… See the full description on the dataset page: https://huggingface.co/datasets/zixiaozhu/MePO.MePO_BPO
MePO Prompt Optimization Dataset (BPO version)
This dataset is designed for research in prompt optimization, particularly for training and evaluating MePO — a lightweight, locally deployable prompt optimization model.
Each JSONL record includes:
rejectedThe original prompt from BP, used as the rejected example.
chosenThe optimized prompt generated by MePO, used as the chosen example.
sliver_responseThe response produced from the BPO prompt (baseline response).
golden_responseThe… See the full description on the dataset page: https://huggingface.co/datasets/zixiaozhu/MePO_BPO.MePO_Alpaca
📦 MePO Prompt Optimization Dataset (Alpaca Version)
The MePO Prompt Optimization Dataset is designed to support research in prompt optimization, especially for training and evaluating MePO — a lightweight and locally deployable prompt optimization model.
📁 Dataset Structure
Each .jsonl record contains the following fields:
rejectedThe original prompt from the Alpaca dataset, serving as the rejected example.
chosenThe optimized prompt generated by MePO, serving as the… See the full description on the dataset page: https://huggingface.co/datasets/zixiaozhu/MePO_Alpaca.Agentic-Chain-of-Thought-Coding-SFT-Dataset-v1.1
🤖 Agentic Coding CoT Dataset v1.1
A high-quality supervised fine-tuning (SFT) dataset for training agentic coding assistants with Chain-of-Thought reasoning capabilities.
📋 Dataset Description
This dataset was created by processing and distilling ~20GB of GitHub crawl data using Minimax-M2 & MiniMax M2.1 to generate structured, reasoning-rich coding examples. Each sample demonstrates systematic problem-solving with explicit tool usage patterns.
🏗️ Assistant… See the full description on the dataset page: https://huggingface.co/datasets/mepartha/Agentic-Chain-of-Thought-Coding-SFT-Dataset-v1.1.