MBPP+

Headlines · 2

1. Oct 20
   TraceForge distills teacher answers into student-style reasoning traces

   A single-author arXiv preprint submitted on 21 October 2026, "TraceForge: Distilling with Your Own Thinking" by Yan Wang, proposes generating a reasoning trace from a teacher model's context and answer in the student model family's own style, instead of recovering hidden reasoning by replaying encrypted blocks — an approach the author says breaks providers' terms of service. Training is two-stage, supervised then adversarial, with a generator writing traces and a critic judging how real a reasoning trace looks. The author reports that an independent judge scored generated traces at 0.68 average authenticity versus 0.66 for real traces and 0.82 for critic-selected traces from four candidates, and that with the same 5,000 questions and teacher answers HumanEval+ improved 3.7 points and MBPP+ 7.1 at the final checkpoint; the paper notes these are authenticity scores, not correctness rates, and the results have not been independently verified.

   Hugging Face · Papers · 🔥 0
2. Oct 14
   AutoSFT: a multi-agent system that automates SFT data research

   A new arXiv paper introduces AutoSFT, a multi-agent system for automated data research in supervised fine-tuning. The authors report that a single agent given the full training stack converged on only one action — adjusting category proportions — while its context filled with execution output and was discarded mid-search. AutoSFT instead uses a meta agent that separates execution from analysis, gathers proposals from independent angles and calls a synthesis subagent for capabilities the data pool cannot cover. The authors report that the executor halves context loss, the analysis subagents lift the best IFEval by 6.5 points, and synthesis adds 3.4 IFEval and 2.9 IFBench points; on code tasks, analysis adds 3.7 MBPP+ over a single agent that does not improve. Every result comes from runs that retrain from the base model each round, with model, trainer and budget fixed within each comparison.

   Hugging Face · Papers · 🔥 0

Experience and discussion from the community

Share my MBPP+ experienceAsk about MBPP+