nbroad/

apigen-with-thinking-1.5k

I added thinking traces to 1.5k samples from Salesforce/APIGen-MT-5k using gpt-oss-120b with high reasoning effort. To prevent HF from modifying the fields, I stored each sample as a json string.

from datasets import load_dataset

ds = load_dataset("nbroad/apigen-with-thinking-1.5k")['train']

from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-14B")

import json

def tokenize(sample): full_ids = tokenizer.apply_chat_template(… See the full description on the dataset page: https://huggingface.co/datasets/nbroad/apigen-with-thinking-1.5k.