QingruZhang / PASTA

PASTA: Post-hoc Attention Steering for LLMs

QingruZhang/PASTA Issues

Memory Usage Increases with Number of Steered Heads
Closed 4 months ago3
I cannot reproduce the demo experiment
Closed 4 months ago3
I cannot run inference
Closed 5 months ago
After adding a space, the model output does not match the requirements at all.
Closed 5 months ago3
Question about the scale_constant in the source code
Closed 6 months ago3
Quickstart code produces unexpected output (same as without PASTA)
Closed 7 months ago2
Question about Efficiency
Closed 7 months ago2