DrRiceIO7/combined-instruct-sft
Combined Instruct SFT Mixture A curated, globally shuffled mixture of 2,980,737 high-quality instruction-following dialogues harmonized into the standardized OpenAI / ChatML format. Dataset Mixture Breakdown Source Dataset Split & Filtering Restrictions Harmonized Count allenai/Dolci-Instruct-SFT Train split, Safety category samples removed 2,041,725 nvidia/Nemotron-SFT-Instruction-Following-Chat-v2 reasoning_off split, English dialogues only 888,460… See the full description on the dataset page: https://huggingface.co/datasets/DrRiceIO7/combined-instruct-sft.
Combined Instruct SFT Mixture
A curated, globally shuffled mixture of 2,980,737 high-quality instruction-following dialogues harmonized into the standardized OpenAI / ChatML format.
Dataset Mixture Breakdown
Format Specification
Every sample is formatted with the standard conversational chat schema:
{
"id": "dolci_personas_math_26ts6mqd3ipqzum75d8qf56e",
"source": "dolci-instruct-sft",
"category": "Math",
"messages": [
{
"role": "user",
"content": "A real estate agent, Emma, encourages financial literacy among her clients..."
},
{
"role": "assistant",
"content": "To determine the impact of post-purchase classes on mortgage management..."
}
],
"metadata": "{\"source_dataset\": \"Tulu 3 Persona MATH\"}"
}Quick Usage
from datasets import load_dataset
dataset = load_dataset("DrRiceIO7/combined-instruct-sft", split="train")
print(dataset)
print(dataset[0])