{ "code": { "dataset": "codeparrot/codeparrot-clean-train", "subset": null, "split": "train", "weight": 0.80, "target_tokens": 3200000000, "note": "Scriptless, immediately streamable permissive-license fallback; Stack-Edu is preferred when Software Heritage access is pre-staged." }, "reasoning": { "dataset": "nvidia/OpenCodeReasoning", "subset": "split_0", "split": "split_0", "weight": 0.15, "target_tokens": 600000000 }, "docs": { "dataset": "HuggingFaceTB/smollm-corpus", "subset": "fineweb-edu-dedup", "split": "train", "weight": 0.05, "target_tokens": 200000000 } }