Update README.md
Browse files
README.md
CHANGED
|
@@ -12,6 +12,8 @@ tags:
|
|
| 12 |
|
| 13 |
--Proof of concept-- in Beta... or theta.
|
| 14 |
|
|
|
|
|
|
|
| 15 |
An experimental approach specifically designed for speech recognition tasks, FAM adapts momentum based on the frequency characteristics of gradient updates.
|
| 16 |
|
| 17 |
### Frequency-Adaptive Momentum (FAM)
|
|
|
|
| 12 |
|
| 13 |
--Proof of concept-- in Beta... or theta.
|
| 14 |
|
| 15 |
+
This optimizer is specifically for ASR type models but works well without the FAM which can be controlled (turned on) by step count. fam_start_step=100.
|
| 16 |
+
|
| 17 |
An experimental approach specifically designed for speech recognition tasks, FAM adapts momentum based on the frequency characteristics of gradient updates.
|
| 18 |
|
| 19 |
### Frequency-Adaptive Momentum (FAM)
|