Attention is a local explorer for seeing what an LLM attends to when comparing source context against markdown query docs: run the extractor to produce a heatmap archive, then inspect token heat, attention-mass source spans, and markdown query items in the browser. Try the hosted explorer at attn.henneberger.dev.
henneberger/attention
Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|