{ "Name": "SmolKalam", "Dialect Subsets": [], "HF Link": "https://huggingface.co/datasets/SultanR/smolkalam", "Link": "https://huggingface.co/datasets/SultanR/smolkalam", "License": "Apache-2.0", "Year": 2025, "Language": "ar", "Dialect": "Modern Standard Arabic", "Source": [ "public datasets" ], "Domain": [ "science", "technology", "general" ], "Form": "text", "Annotation Style": [ "machine annotation", "LLM annotation" ], "Description": "Large-scale Arabic SFT dataset with reasoning and tool-calling.", "Volume": 1790478.0, "Unit": "sentences", "Ethical Risks": "Low", "Provider": [ "KAUST" ], "Derived From": [ "Smoltalk2", "OpenThoughts" ], "Paper Title": "SmolKalam: Ensemble Quality-Filtered Translation at Scale for High Quality Arabic Post-Training Data", "Paper Link": "https://arxiv.org/pdf/2511.18411.pdf", "Script": "Arab", "Tokenized": false, "Host": "HuggingFace", "Access": "Free", "Cost": "", "Has Splits": false, "Partial": false, "Tasks": [ "instruction tuning" ], "Venue Title": "arXiv", "Venue Type": "preprint", "Venue Name": "arXiv", "Authors": [ "Sultan AlRashed", "Chadi Helwe", "Francesco Orabona" ], "Affiliations": [], "Abstract": "Although the community has tackled the acquisition of high-quality Arabic pre-training data, we still lack large-scale, multi-turn Arabic datasets that include reasoning and tool calling. Naive translation can work at the pretraining scale, but post-training demands much higher quality, which requires a stricter approach to dataset curation. In this work, we introduce SmolKalam, a translation of Smoltalk2 that uses a multi-model ensemble translation pipeline, applies quality filtering, and examines effective translation techniques for traditional decoder-only models through ablations.", "Added By": "Zaid Alyafeai" }