Shared KV caching across the last 25% to 50% of layers, adapted incrementally from pretrained LLaMA weights, is claimed to improve training speed and inference throughput with little accuracy loss.
https://huggingface.co/TinyLlama/TinyLlama_v1.1
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
ECHO-LLaMA: Efficient Caching for High-Performance LLaMA Training
Shared KV caching across the last 25% to 50% of layers, adapted incrementally from pretrained LLaMA weights, is claimed to improve training speed and inference throughput with little accuracy loss.