Under identical MLP-only pruning schedules, KL self-distillation matches or exceeds cross-entropy fine-tuning for recovering test accuracy of a pruned 7B language model, with a 2.5 to 5.5 point edge at 50% retention.
EPSD: Early pruning with self-distillation for efficient model compression, 2024
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Constrained Edge AI Deployment: Fine-Tuning vs Distillation for LLM Compression
Under identical MLP-only pruning schedules, KL self-distillation matches or exceeds cross-entropy fine-tuning for recovering test accuracy of a pruned 7B language model, with a 2.5 to 5.5 point edge at 50% retention.