A greedy, ablation-based pruning method extracts a standalone task-specific subnetwork from GPT-2 Small, reducing parameters by up to 82.77% while keeping accuracy on three synthetic single-token tasks.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Extracting Interpretable Task-Specific Circuits from Large Language Models for Faster Inference
A greedy, ablation-based pruning method extracts a standalone task-specific subnetwork from GPT-2 Small, reducing parameters by up to 82.77% while keeping accuracy on three synthetic single-token tasks.