This page provides a summary of all released models. Download links are also included. We will continue to release improved models and new architectures.
Description: For the bias-scale-norm tuned models, a learnable channel-wise scaling factor is added before the fully connected layers, and a learnable channel-wise bias is added after the fully connected layers. The parameters in the normalization layers are also tuned.
| Version | Training data | Download url | Notes |
|---|---|---|---|
| v1 | ShareGPT | Google drive | Initial version |
Coming soon