heihei/

prune-to-prosper-data

This dataset contains experimental data for the paper "Dimensions Are Interchangeable: Evidence That Task-Aware Embedding Pruning Does Not Outperform Random Selection".

	Dataset Structure




	
	analyze/ — Per-Model Chunk Importance Analysis

Chunk-level importance scores for 3 models evaluated with win_size=2 (512 chunks for 1024-dim models):

File Model Size

gte-large-en-v1.5.json GTE-Large 4.7 MB… See the full description on the dataset page: https://huggingface.co/datasets/heihei/prune-to-prosper-data.