transformers
768dc93a
- H-Net: ponytail cuts
Go
Login via GitHub
Home
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
22 days ago
H-Net: ponytail cuts Drop num_key_value_heads and use_mamba_kernels (Zamba2/NemotronH configs now warn instead of the mixer), inline the window mask, simpler state restore and renames.
References
#50167 - Add H-Net
Author
ArthurZucker
Parents
3d19e6a4
Loading