Downloads · 30 days
0
Luki-G/augmented-residuals-culturax24
augmented-residuals-culturax24 is a machine learning model from Luki-G. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Backup of the corrected AttnRes (attnress1) pretraining trajectory and available seed-42 full fine-tuning checkpoints. Earlier files in this repository are retained.
Downloads · 30 days
0
Access
Public
Updated Oct 27, 2026
Repo size
148 GB
Likes
0
Public
Click a slice to open those files.
.pt121 GB · 100%
From the Hugging Face model README
Backup of the corrected AttnRes (attnres_s1) pretraining trajectory and available seed-42 full fine-tuning checkpoints. Earlier files in this repository are retained.
pretrained/attnres_s1/ contains complete checkpoint files from the corrected October 2026 run. Steps 75,000 and 85,000 had only incomplete scratch archive files and are excluded. The final checkpoint is step 115,200.
finetuned/*/{xnli,pawsx}/seed42/model.pt contains model_state_dict and a training summary. finetuned/attnres_s1/xquad/seed42_frozen_routing/model.pt contains model, pretrain_iteration, best_epoch, and freeze_routing; its result and training summaries are alongside it. PAWS-X frozen-routing runs saved summaries but did not save model weights.
The tokenizer used by these models is in tokenizer/. See backup_manifest.json for paths included in this backup.