Unverified50% confidenceFactExact time
Compared to MHA, GQA can compress the KV cache size to one-quarter without noticeably degrading model quality.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMLA theoretically reduces the GPU memory footprint of KV Cache to 5% to 13% of standard MHA by compressing KV matrices into low-dimensional latent vector caches.79% similarUnverifiedThe 70B version of LLaMA 2 uses 8-group GQA, reducing the KV cache to 1/4 of MHA while barely affecting model quality.73% similarUnverifiedGQA groups attention heads to share KV projections, which can compress the KV cache to 1/4 or less of the original size with almost no loss in model quality.71% similarUnverifiedTechniques such as GQA and MLA can significantly compress KV cache size; DeepSeek V3 adopts MLA to reduce memory overhead in long-context scenarios.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/915869API
curl https://kongchang.com/api/v1/knowledge/claims/915869MCP
get_claim(id=915869)