Knowledge distillation from GPT-3.5 into Llama3.1-8B improves bundle-generation precision and coverage but not recall, and the utilization method (SFT vs ICL) is the strongest determinant.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Does Knowledge Distillation Matter for Large Language Model based Bundle Generation?
Knowledge distillation from GPT-3.5 into Llama3.1-8B improves bundle-generation precision and coverage but not recall, and the utilization method (SFT vs ICL) is the strongest determinant.