SupraLabs releases a 5M-row reasoning corpus for tiny-model fine-tuning

LH-Tech_AI · reddit · 2026-07-24

SupraLabs released reasoning-corpus-4K-5M-v1, a 5-million-row dataset aimed at training tiny SLMs to reason.

• Each sample includes repoid, token length, user prompt, thoughttrace, assistant answer, and a ChatML-formatted record.
• All examples are capped at 5k sequence length to make them suitable for SFT and fine-tuning.
• The dataset is published on Hugging Face and the post invites feedback, follows, and reuse.
• The release is positioned as a large reasoning corpus for small-model training rather than a benchmark or model update.

Original post →

More from Research

• Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
• Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
• FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
• LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26
• Behind the EMNLP keynote: LMs as simulated learners to generate child language acquisition hypotheses — najoungkim · 2026-10-26
• The paper behind the EMNLP 2026 keynote on bigram 'simulated learners' — najoungkim · 2026-10-26

Research channel →