Compressing the KV cache during prefilling as well as decoding enables larger batch sizes and roughly 44% higher LLM inference throughput at near-full accuracy in long-context settings.
N., Çaglar G \"u lçehre, and Xiang, B
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
-
Batch-Max: Higher LLM Throughput using Larger Batch Sizes and KV Cache Compression
Compressing the KV cache during prefilling as well as decoding enables larger batch sizes and roughly 44% higher LLM inference throughput at near-full accuracy in long-context settings.