specific method · filed under model architecture
Block AttnRes
A block-based form of attention residuals that partitions layers into blocks to reduce overhead.
Also called block attention residual, Block Attention Residuals.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- core
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1core 1
Documented in
Evidence
2 spans quoted from the sources, strongest treatment first.
For the attention residual, we design a companion optimization based on Block AttnRes.
coremodel architecturein Kimi K3Moonshot AI
To reduce this overhead, we partition the L layers into N blocks of S=L/N layers each.
usedmodel architecturein Kimi K3Moonshot AI
Filed alongside
Other methods under model architecture :: normalization & residual.
Gated ResidualManifold-Constrained Hyper-ConnectionsAttention Residuals (AttnRes)GatedNormIdentity Hyper-ConnectionsRMSNormSingle-Pass mHCHyper-ConnectionsIndependent Per-Branch NormalizationPer-Branch Scalar Write GateQK-ClipAll-branch Elementwise Read GateBounded Positive GatesData-Dependent Residual Read and Write OperatorsDynamic Gating of Residual Reads and WritesElementwise Data-Dependent Residual Read GateFour-Branch Residual StreamFour-Stream Hyper-Connection Combine UpdateFull Attention ResidualsLow-Rank Gated MixPost-Embedding RMSNormPre-LNSimplified AltUpSparse Gated Residual Writes