update: recipes

Author: kozistr

tests/constants.py

@@ -310,16 +310,14 @@
     (Adan, {'lr': 5e-1, 'max_grad_norm': 1.0}, 5),
     (Adan, {'lr': 5e-1, 'weight_decay': 1e-3, 'use_gc': True}, 5),
     (Adan, {'lr': 5e-1, 'weight_decay': 1e-3, 'weight_decouple': True}, 5),
-    (DAdaptAdaGrad, {'lr': 2.0, 'weight_decay': 1e-3}, 50),
-    (DAdaptAdaGrad, {'lr': 2.0, 'weight_decay': 1e-3, 'momentum': 0.1}, 50),
-    (DAdaptAdam, {'lr': 2.0, 'weight_decay': 1e-3}, 25),
-    (DAdaptAdam, {'lr': 1.0, 'weight_decay': 1e-3, 'weight_decouple': True}, 50),
-    (DAdaptSGD, {'lr': 2.0, 'weight_decay': 1e-2}, 25),
-    (DAdaptSGD, {'lr': 2.0, 'momentum': 0.9, 'weight_decay': 1e-3}, 25),
-    (DAdaptAdan, {'lr': 1.0, 'weight_decay': 1e-2}, 25),
-    (DAdaptAdan, {'lr': 1.0, 'weight_decay': 1e-2, 'weight_decouple': True}, 50),
-    (AdamS, {'lr': 1.0, 'weight_decay': 1e-3}, 10),
-    (AdamS, {'lr': 1.0, 'weight_decay': 1e-3, 'ams_bound': True}, 20),
+    (DAdaptAdaGrad, {'lr': 2e0, 'weight_decay': 1e-3}, 50),
+    (DAdaptAdaGrad, {'lr': 2e0, 'weight_decay': 1e-3, 'momentum': 0.1}, 50),
+    (DAdaptAdam, {'lr': 5e2, 'weight_decay': 1e-3}, 25),
+    (DAdaptSGD, {'lr': 2e0, 'weight_decay': 1e-3}, 25),
+    (DAdaptAdan, {'lr': 1e0, 'weight_decay': 1e-2}, 25),
+    (DAdaptAdan, {'lr': 1e0, 'weight_decay': 1e-2, 'weight_decouple': True}, 50),
+    (AdamS, {'lr': 1e0, 'weight_decay': 1e-3}, 10),
+    (AdamS, {'lr': 1e0, 'weight_decay': 1e-3, 'ams_bound': True}, 20),
     (AdaFactor, {'lr': 7.5e-1, 'weight_decay': 1e-3, 'scale_parameter': False}, 100),
     (AdaFactor, {'lr': 7.5e-1, 'weight_decay': 1e-3, 'ams_bound': True}, 125),
     (Apollo, {'lr': 5e-1, 'weight_decay': 1e-3}, 10),