Synthetic-document fine-tuning on Qwen2.5-7B: a controlled study of whether SDF installs sandbagging, finding a layered recognition/generation/behavior dissociation.
-
Updated
May 25, 2026 - Python
Synthetic-document fine-tuning on Qwen2.5-7B: a controlled study of whether SDF installs sandbagging, finding a layered recognition/generation/behavior dissociation.
AI-safety research: inducing CoT unfaithfulness in DeepSeek-R1-Distill-Qwen-7B via Synthetic Document Finetuning. SDF installed a conditional 'sleeper' backdoor (accuracy 69%→18%, hint-following 16%→31% when triggered) but not covert behavioral transfer.
Add a description, image, and links to the synthetic-document-finetuning topic page so that developers can more easily learn about it.
To associate your repository with the synthetic-document-finetuning topic, visit your repo's landing page and select "manage topics."