Channel Merging clusters similar channel parameters from fine-tuned LLMs and reconstructs the selected expert at inference, matching unmerged accuracy with about 53% of the ensemble's parameters.
Chinese-Mistral-7B-Instruct-v0.1
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Channel Merging: Preserving Specialization for Merged Experts
Channel Merging clusters similar channel parameters from fine-tuned LLMs and reconstructs the selected expert at inference, matching unmerged accuracy with about 53% of the ensemble's parameters.