BashCache/temp-llada-base-20-snr-weighted
7B • Updated • 57
GSAI-ML/LLaDA-8B-Base structurally pruned with KFAC-OBD (SNR-weighted) at 20/30/40% sparsity, with LoRA adapters.
Note 20% structured-pruned LLaDA-8B-Base (KFAC-OBD, SNR-weighted)
Note LoRA adapter recovering the 20% pruned model; load on top of BashCache/temp-llada-base-20-snr-weighted
Note 30% structured-pruned LLaDA-8B-Base (KFAC-OBD, SNR-weighted)
Note LoRA adapter recovering the 30% pruned model; load on top of BashCache/temp-llada-base-30-snr-weighted
Note 40% structured-pruned LLaDA-8B-Base (KFAC-OBD, SNR-weighted)
Note LoRA adapter recovering the 40% pruned model; load on top of BashCache/temp-llada-base-40-snr-weighted