LGCLJun 15

QK-Normed MLA: QK normalization without full key caching

arXiv:2606.163106.6
Predicted impact top 65% in LG · last 90 daysOriginality Incremental advance
AI Analysis

For practitioners training large language models with MLA, this provides a practical way to use QK normalization for improved stability and performance without sacrificing decoding efficiency.

The paper resolves the apparent incompatibility between QK normalization and Multi-head Latent Attention (MLA) by showing that QK RMSNorm can be implemented without full key caching, achieving lower training loss and better downstream accuracy than QK clipping with less than 2% latency overhead up to 256k context.

Query-key (QK) normalization stabilizes attention by controlling the scale of queries and keys before the dot product, but is not immediately compatible with Multi-head Latent Attention (MLA). MLA achieves efficient decoding by caching low-dimensional latent states instead of full keys, whereas post-projection QK RMSNorm appears to require the fully projected key for every cached token. We show this apparent incompatibility is an implementation artifact, not an architectural constraint. RMSNorm decomposes into a static affine weight and a dynamic scalar RMS statistic. The static key-side weight can be absorbed into the MLA query-side projection; the dynamic key statistic reduces to one inverse-RMS scalar per token and KV group. The resulting formulation is exactly equivalent to explicit post-projection QK RMSNorm in exact arithmetic and preserves MLA's latent decode path. In our 400M runs trained for up to 100B tokens, QK-Normed MLA achieves lower training loss and better downstream accuracy than QK clipping, while H800 decode benchmarks show less than 2% latency overhead up to 256k context. These results make QK normalization a practical stabilization option for MLA models without requiring full-key caching.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes