rStar-Coder SFT Subset

← All datasets

Author: hf-mirror · License: cc-by-4.0 · Format: parquet

0 downloads · 0 stars · 0 mirrors · 12,079,419,204 · 20 files

Microsoft rStar-Coder SFT slice: ~11 GB of verified competitive-coding + code-reasoning problems (CC-BY-4.0).

Tags: coding · reasoning · sft · microsoft · competitive-programming

Description

Microsoft's rStar-Coder SFT subset (seed_sft split, 20 parquet shards, ~11 GB): high-difficulty competitive-coding and general code reasoning problems with verified solutions, assembled for the rStar-Coder reasoning recipe. The full repo also ships massive RL query/reward corpora (~473 GB total) — this pins the trainable SFT slice. CC-BY-4.0.

Download

Direct download: https://huggingface.co/datasets/microsoft/rStar-Coder/resolve/main/seed_sft/data-00000-of-00020.parquet

Or via the API: curl -L -o file "https://huggingface.co/datasets/microsoft/rStar-Coder/resolve/main/seed_sft/data-00000-of-00020.parquet"

Files

• seed_sft/data-00000-of-00020.parquet (574,933,145)
• seed_sft/data-00001-of-00020.parquet (624,317,720)
• seed_sft/data-00002-of-00020.parquet (653,308,209)
• seed_sft/data-00003-of-00020.parquet (527,739,573)
• seed_sft/data-00004-of-00020.parquet (647,820,731)
• seed_sft/data-00005-of-00020.parquet (374,035,020)
• seed_sft/data-00006-of-00020.parquet (385,620,188)
• seed_sft/data-00007-of-00020.parquet (585,829,649)
• seed_sft/data-00008-of-00020.parquet (687,414,595)
• seed_sft/data-00009-of-00020.parquet (640,157,828)
• seed_sft/data-00010-of-00020.parquet (616,548,103)
• seed_sft/data-00011-of-00020.parquet (628,288,025)
• seed_sft/data-00012-of-00020.parquet (602,893,990)
• seed_sft/data-00013-of-00020.parquet (625,303,434)
• seed_sft/data-00014-of-00020.parquet (662,642,293)
• seed_sft/data-00015-of-00020.parquet (681,419,592)
• seed_sft/data-00016-of-00020.parquet (583,723,330)
• seed_sft/data-00017-of-00020.parquet (744,966,704)
• seed_sft/data-00018-of-00020.parquet (638,507,353)
• seed_sft/data-00019-of-00020.parquet (593,949,722)

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