Add GLM-5 (754B MoE) contrib model for trn2.48xlarge#143

Open

jimburtoft wants to merge 10 commits into

aws-neuron:mainfrom

jimburtoft:contrib/GLM-5

Commits on Apr 24, 2026

Add GLM-5 (754B MoE) contrib model for NxDI on trn2.48xlarge
jimburtoft
committed

Commits on Apr 25, 2026

Enable NKI MLP kernel for GLM-5 dense layers (0-2), +4% throughput
jimburtoft
committed

Commits on Apr 26, 2026

Update GLM-5 contrib: verified benchmarks, NKI 0.3.0 nkilib ref, fix docs
jimburtoft
committed
Remove NKI kernel impact comparison table from README
jimburtoft
committed
Restore NKI kernel impact table (Neuron-only, no GPU comparisons)
jimburtoft
committed

Commits on Apr 30, 2026

Switch GLM-5 MoE from fused NKI mega-kernel to non-fused ExpertMLPsV2 path
jimburtoft
committed

Commits on May 1, 2026

Revert non-fused ExpertMLPsV2 path: 17% slower than fused NKI kernel
jimburtoft
committed

Commits on May 2, 2026

Support reduced num_hidden_layers for profiling (remove excess layer weights)
jimburtoft
committed

Commits on May 23, 2026

Add MLA attention NKI kernel for GLM-5 TKG decode (cos_sim=0.999996)
jimburtoft
committed

Commits on May 26, 2026

Optimize MLA NKI kernel: hoist DMA loads, reduce masking ops, simplify normalization
jimburtoft
committed