self-training
SII_self_evovling_02_training_datasetself-training-denoising-forgetting
Self-Training: Denoising vs. Forgetting — Reproduction Artifacts
Reproduction artifacts for "Why Self-Training Helps and Hurts" (arXiv:2602.14029):
the U-shaped risk curve from the denoising–forgetting trade-off, in overparameterized
linear regression (synthetic, spiked covariance) and a small CIFAR-10 deep-learning analogue.
Read REPORT.md first — it states which claims reproduced, with what numbers,
and lists all assumptions, deviations, hardware and cost.
Contents… See the full description on the dataset page: https://huggingface.co/datasets/pngwn/self-training-denoising-forgetting.SI2CA-Training-TrajectoriesDataset Card for SI2CA-Training-Trajectories
[🌐 Website] •
[🤗 Dataset] •
[📜 Paper] •
[🐱 GitHub]
💡 Introduction
This dataset consists of 32,340 coding-agent trajectories generated by Qwen3.5-122B-A10B on the same 10,780 executable Python SWE tasks under the three trajectory-curation settings of Section 4.4 of the paper: standard sampling, full self-judgement, and an efficient discovered strategy found by the recursive self-improvement framework. Each task is… See the full description on the dataset page: https://huggingface.co/datasets/Self-Improving-Coding-Agents/SI2CA-Training-Trajectories.Self_images_Training_Datasetgpt4-self-instructself_training_implementations_250.jsonl