Datasets:
|
Download README.md from OpenIntelligenceNet/Sanitized-Dataset-SFT: direct link, hf CLI and curl.
- Browser
- Download file 1.55 kB
-
https://huggingface.co/datasets/OpenIntelligenceNet/Sanitized-Dataset-SFT/resolve/main/README.md
- Command line
-
hf download hf://datasets/OpenIntelligenceNet/Sanitized-Dataset-SFT/README.md
-
curl -L -o README.md https://huggingface.co/datasets/OpenIntelligenceNet/Sanitized-Dataset-SFT/resolve/main/README.md
1.55 kB
metadata
dataset_info:
features:
- name: source
dtype: string
- name: messages
list:
- name: role
dtype: string
- name: content
dtype: string
- name: thinking
dtype: string
- name: text
dtype: string
splits:
- name: train
num_examples: 84764
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
task_categories:
- text-generation
language:
- en
tags:
- minicpm-5
- reasoning
- uncensored
- chatml
Sanitized SFT Dataset (85k)
This dataset contains exactly 84,764 rows optimized for MiniCPM-5 Supervised Fine-Tuning (SFT) and Importance Matrix (Imatrix) calibration.
Composition
- Uncensored (28k): High-quality, zero-refusal conversational data pulled from Alpaca-Uncensored and ShareGPT unfiltered variants.
- Reasoning & Factual (57k): Logic, math, and code-heavy data pulled from DeepSeek-v4, Claude 3.5, MetaMath, Hunter-Alpha, and Kimi-Cyber-Reasoning.
Formatting
- The
textcolumn contains pre-rendered strings using the strict MiniCPM-5 Jinja ChatML format. - Sanitized: Internal literal tags (like
<|im_end|>or<think>discussed in coding problems) have been escaped to prevent tokenizer poisoning during training. - Native
<think>blocks are enforced for every assistant turn (populated if reasoning exists, explicitly empty if not).
Usage
Perfect for direct loss-masking during SFT or passing to llama.cpp for accurate quantization calibration.