diff --git a/configs/context_numbers_128.yaml b/configs/context_numbers_128.yaml
index 37e29eb..e7374a6 100644
--- a/configs/context_numbers_128.yaml
+++ b/configs/context_numbers_128.yaml
@@ -20,7 +20,7 @@ per_device_train_batch_size: 64
 per_device_eval_batch_size: 64
 max_val_samples_per_ds: 50
 # optim: schedule_free_adamw
-learning_rate: 0.0001
+learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
 weight_decay: 0.1
diff --git a/configs/context_numbers_256.yaml b/configs/context_numbers_256.yaml
index 3e639b1..4c34100 100644
--- a/configs/context_numbers_256.yaml
+++ b/configs/context_numbers_256.yaml
@@ -20,7 +20,7 @@ per_device_train_batch_size: 32
 per_device_eval_batch_size: 32
 max_val_samples_per_ds: 20
 # optim: schedule_free_adamw
-learning_rate: 0.0001
+learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
 weight_decay: 0.1
diff --git a/configs/pwc.yaml b/configs/pwc.yaml
index b4b00e1..535d527 100644
--- a/configs/pwc.yaml
+++ b/configs/pwc.yaml
@@ -20,7 +20,7 @@ per_device_train_batch_size: 32
 per_device_eval_batch_size: 32
 max_val_samples_per_ds: 20
 # optim: schedule_free_adamw
-learning_rate: 0.0001
+learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
 weight_decay: 0.1
diff --git a/configs/pwc_and_ctx_numbers_256.yaml b/configs/pwc_and_ctx_numbers_256.yaml
index 4046e06..2fe76c8 100644
--- a/configs/pwc_and_ctx_numbers_256.yaml
+++ b/configs/pwc_and_ctx_numbers_256.yaml
@@ -20,7 +20,7 @@ per_device_train_batch_size: 32
 per_device_eval_batch_size: 32
 max_val_samples_per_ds: 20
 # optim: schedule_free_adamw
-learning_rate: 0.0001
+learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
 weight_decay: 0.1