transformers
431e9e9d
- fix(gemma3n): Clamp the AltUp magnitudes instead of using a CPU epsilon tensor (#50457)
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
5 days ago
fix(gemma3n): Clamp the AltUp magnitudes instead of using a CPU epsilon tensor (#50457) fix: Clamp the Gemma 3n altup magnitudes instead of using a CPU epsilon tensor Signed-off-by: Harshal Janjani <harshaljanjani@gmail.com>
References
#50457 - fix(gemma3n): Clamp the AltUp magnitudes instead of using a CPU epsilon tensor
Author
harshaljanjani
Parents
de35c2b1
Loading