MGAA allocates low-rank compression budget by sublayer input-output cosine similarity and by matrix energy retention, improving compressed LLM perplexity and reasoning accuracy.
Awq: Activation-aware weight quantization for llm compression and acceleration,
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
-
MGAA: Multi-Granular Adaptive Allocation fof Low-Rank Compression of LLMs
MGAA allocates low-rank compression budget by sublayer input-output cosine similarity and by matrix energy retention, improving compressed LLM perplexity and reasoning accuracy.