rot13 seq-QA LoRA rank-1 (alpha2) task-vector (avg-of-6) - held-out receptor software

Delta = mean_6( fold(LoRA seqqa_rot13) - docsonly_rot13 ), sources people+planets x seeds 1/2/3, LR=1e-4. Apply: W(hugo/v1-7b-software-docsonly-seed[1-3]) + lambda*Delta. fp32. LoRA sweet-spot lambda~2.

Downloads last month

-

Downloads are not tracked for this model. How to track
Safetensors
Model size
8B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support