---
license: openrail
---
Llama-2 representations extracted from Alpaca instruction tuning dataset (original orderof examples is preserved). Representations are extracted from the final layer and averaged across all tokens.

Dataset structure: key=sample id, value=feature vector in string format, with ";" separator.