Author: hf-mirror · License: non-commercial research only (mixed upstream: CC BY-NC + Alibaba Cloud Model Studio terms) · Format: parquet
0 downloads · 0 stars · 0 mirrors · 42,577,138 · 6 files
57,937 filtered multi-teacher SFT traces from Qwen3.8-Max, GLM-5.2, and Kimi K3 across 24 configs with per-row provenance.
Tags: sft · distilled · multi-teacher · reasoning · tool-calling · multilingual · non-commercial
Quality-filtered, deduplicated multi-teacher SFT corpus of 57,937 traces: 48,283 from Qwen3.8-Max-Preview, 5,307 from GLM-5.2, and 4,347 from Kimi Code K3, across math, code, reasoning, instruction-following, tool use, science, long-context, multilingual, and creative dialogue. Ships 24 configs (sft_balanced default, canonical audit view with per-row provenance columns, openai_messages, and more) in six languages. NON-COMMERCIAL: the repo's LICENSE limits use to controlled noncommercial research — the mixture combines CC BY-NC-SA 4.0 / CC BY-NC 3.0 sources with Alibaba Cloud Model Studio terms (§4.48(d)(v) competitive-use restriction on Qwen3.8-Max outputs); per-row source licenses are retained in the canonical config.
Direct download: https://huggingface.co/datasets/r0b0tlab/qwen3.8-max-glm5.2-kimi-k3-distillation/resolve/main/data/sft_balanced/train-00000-of-00006.parquet
Or via the API: curl -L -o file "https://huggingface.co/datasets/r0b0tlab/qwen3.8-max-glm5.2-kimi-k3-distillation/resolve/main/data/sft_balanced/train-00000-of-00006.parquet"
Added 2026-09-17. Platform: Distillio — the open source training data collective.