Does MLA losslessly compress an existing transformer cache?
No. MLA is a learned model architecture. DeepSeek’s reported reduction is relative to its stated baseline; it is not a generic lossless compressor for any pretrained model.
Answered in
The KV Cache: How GQA and DeepSeek MLA Reduce Inference MemoryKV-cache size depends on layers, heads, precision and context. GQA shares key-value heads; MLA caches a learned latent plus positional data.
Read the full analysisOther questions this article answers
More how ai actually works questions
- What does Google's Deep Review accept?
- Did the Paper Assistant Tool replace conference peer review?
- Are uploaded documents used to train Google models?
- When does Google Workspace begin enforcing the MCP OAuth client marker?
- Will this change break regular Google Workspace REST API integrations?
- What should a team do if one OAuth client serves agents and regular integrations?
- What is NVIDIA Dory 1.0?
- Does increasing Dory's loop count make it a larger model?
Every answer on Crashtech is written by the editor of the article it comes from — never auto-summarised. Browse all answers or the How AI Actually Works beat.