Hparams:

  • LR: 1e-5
  • Weight Decay: 0.02
  • grad-clip: 0.2
  • R64, A32, Rslora, 0.1 drop out, 16bit lora (qlora is cope.)

pretrained on a buncha LNs, books, etc. Used for a base for Pascal.

Downloads last month
45
Safetensors
Model size
12.2B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Delta-Vector/Pascal-Base-12B

Merge model
this model
Finetunes
1 model
Quantizations
4 models

Collection including Delta-Vector/Pascal-Base-12B