DeepSeek-V2
DeepSeek-V2 — MLA (Multi-head Latent Attention)
Why it matters
A third path, distinct from MQA and GQA, that compresses KV into a low-dimensional latent
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2405.04434Added in Week 35, 2026 · See that week