Unverified50% confidenceFactExact time
GQA 通过让多个 Query 头共享同一组 Key-Value 头，在保持模型表达力的同时降低了 KV Cache 的显存占用
1
Sources
50%
Confidence
Long-term
Relevance
8/25/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedLLaMA 3 8B adopts Grouped Query Attention (GQA), reducing the number of Key and Value attention heads to 1/4 of the number of Query heads.68% similarUnverifiedMuse Glimmer uses Grouped Query Attention (GQA) with only 2 KV heads, making the KV Cache extremely memory-efficient.66% similarUnverifiedKV cache reuses the Key and Value matrices of historical tokens during the inference phase, avoiding repeated attention computation on already-generated sequences.65% similarVerifiedThe self-attention mechanism uses dot-product operations among three matrices—Query, Key, and Value—to overcome the parallelization bottleneck caused by the sequential computation of RNNs/LSTMs.63% similarUnverifiedGQA (Grouped Query Attention) was introduced by Meta in LLaMA 257% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/798527API
curl https://kongchang.com/api/v1/knowledge/claims/798527MCP
get_claim(id=798527)