gla_100M_r64_bs256_lr3e-4_steps5000

GLA (Gated Linear Attention) 100M (rank 64) — Low-rank Fast-Weight Ablation

Pretrained 100M-parameter GLA (Gated Linear Attention) with low-rank parameterization (r64) on FineWeb-Edu. Part of a 16-cell ablation (4 archs × 4 ranks: r8, r32, r64, rfull) studying whether constraining the q/k/v fast-weight projections (or LaCT's SwiGLU MLP) to low rank can match or exceed full-rank performance.

Training

Code: see run_main_100M.sh.

Eval results

• FineWeb-Edu val PPL: 23.80
• MQAR (multi-query associative recall):
  • K=4: 0.115
  • K=16: 0.581
  • K=64: 0.763
  • K=256: 0.864
• LAMBADA acc: 0.114
• HellaSwag acc_norm: 0.279
• ARC-Easy acc_norm: 0.403
• PIQA acc_norm: 0.579
• WinoGrande acc: 0.500

Notes

• This is one of 16 cells; the other rank/arch combinations are uploaded under the same HF org (nlproj) with repo names matching the local dump folder, e.g. nlproj/gla_100M_{r8|r32|r64|rfull}_bs256_lr3e-4_steps5000.
• Key finding of the ablation: at this scale, low rank often matches or beats full rank on downstream tasks (LoRA-style "adaptation is intrinsically low-rank" hypothesis). GatedDeltaNet is the exception — its rfull is the strongest in the whole sweep on PPL / LAMBADA / HellaSwag / ARC-Easy.

Run name: gla_100M_r64_bs256_lr3e-4_steps5000