Magicoder OSS-Instruct 75K (decontaminated)

← All datasets

Author: hf-mirror · License: mit · Format: jsonl

0 downloads · 0 stars · 0 mirrors · 203,201,169 · 1 file

75k GitHub-seeded, execution-validated code instruction pairs (Magicoder recipe, decontaminated, MIT).

Tags: coding · sft · magicoder · uiuc

Description

The Magicoder OSS-Instruct corpus (UIUC): 75k instruction-response pairs generated by open-source LLMs seeded from 13k real GitHub code snippets, then execution-validated and decontaminated against HumanEval/MBPP — the recipe behind the Magicoder models. MIT.

Download

Direct download: https://huggingface.co/datasets/ise-uiuc/Magicoder-OSS-Instruct-75K/resolve/main/data-oss_instruct-decontaminated.jsonl

Or via the API: curl -L -o file "https://huggingface.co/datasets/ise-uiuc/Magicoder-OSS-Instruct-75K/resolve/main/data-oss_instruct-decontaminated.jsonl"

Files

• data-oss_instruct-decontaminated.jsonl (203,201,169)

Added 2026-09-27. Platform: Distillio — the open source training data collective.