ContextEcho — Released Dataset Per cell evaluation corpus and donated session prefixes for the ContextEcho benchmark. This Hugging Face repository hosts the released dataset artifacts. The canonical project page, latest README, code, reproduction instructions, and donation workflow are maintained on GitHub: https://github.com/Accenture/ContextEcho For the formal datasheet, see DATASHEET.md . What's here Total: 41,924 files, ~1 GB. Note: the donated session transcripts under data/sessions/ .jsonl remain available from the Files and versions tab. They are intentionally not exposed as a Hugging Face Dataset Viewer split because raw long session agent logs can contain escape patterns that the web viewer's streaming JSON parser rejects. Download those files directly instead. Quick start Reading a single session prefix The first turn of each session is a permission mode metadata record; the conversation begins on turn 2. Reading a per cell evaluation Loading the manifest Reproducing a benchmark figure To reproduce the headline drift gap forest plot, use the public GitHub code repository together with this dataset's results/ directory: For current reproduction notes, see the GitHub README…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy