diff --git a/configs/context_numbers_10.yaml b/configs/context_numbers_10.yaml
index 1ed026b..c29d1ac 100644
--- a/configs/context_numbers_10.yaml
+++ b/configs/context_numbers_10.yaml
@@ -27,8 +27,7 @@ max_val_samples_per_ds: 500
 learning_rate: 0.0001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.05
+weight_decay: 0.01
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_128.yaml b/configs/context_numbers_128.yaml
index fed257d..bf7a88d 100644
--- a/configs/context_numbers_128.yaml
+++ b/configs/context_numbers_128.yaml
@@ -26,8 +26,8 @@ max_val_samples_per_ds: 20
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.05
+weight_decay: 0.01
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_128_only.yaml b/configs/context_numbers_128_only.yaml
index 28a5f6f..526bdf4 100644
--- a/configs/context_numbers_128_only.yaml
+++ b/configs/context_numbers_128_only.yaml
@@ -26,8 +26,8 @@ max_val_samples_per_ds: 20
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.05
+weight_decay: 0.01
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_256.yaml b/configs/context_numbers_256.yaml
index ca48999..66ece09 100644
--- a/configs/context_numbers_256.yaml
+++ b/configs/context_numbers_256.yaml
@@ -26,8 +26,8 @@ max_val_samples_per_ds: 20
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.05
+weight_decay: 0.01
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_256_only.yaml b/configs/context_numbers_256_only.yaml
index 47ce81c..68489c6 100644
--- a/configs/context_numbers_256_only.yaml
+++ b/configs/context_numbers_256_only.yaml
@@ -26,8 +26,8 @@ max_val_samples_per_ds: 20
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.05
+weight_decay: 0.01
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_32.yaml b/configs/context_numbers_32.yaml
index f39138d..61771db 100644
--- a/configs/context_numbers_32.yaml
+++ b/configs/context_numbers_32.yaml
@@ -26,8 +26,7 @@ max_val_samples_per_ds: 20
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.05
+weight_decay: 0.01
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_32_only.yaml b/configs/context_numbers_32_only.yaml
index 5cc0c73..73e2fd7 100644
--- a/configs/context_numbers_32_only.yaml
+++ b/configs/context_numbers_32_only.yaml
@@ -26,8 +26,7 @@ max_val_samples_per_ds: 20
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.05
+weight_decay: 0.01
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_64.yaml b/configs/context_numbers_64.yaml
index 4c69eaf..d4bfc03 100644
--- a/configs/context_numbers_64.yaml
+++ b/configs/context_numbers_64.yaml
@@ -26,8 +26,8 @@ max_val_samples_per_ds: 20
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.05
+weight_decay: 0.01
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_64_only.yaml b/configs/context_numbers_64_only.yaml
index 42f6bfe..dc76775 100644
--- a/configs/context_numbers_64_only.yaml
+++ b/configs/context_numbers_64_only.yaml
@@ -26,8 +26,8 @@ max_val_samples_per_ds: 20
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.05
+weight_decay: 0.01
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_debug.yaml b/configs/context_numbers_debug.yaml
index 9a5827d..64ef35b 100644
--- a/configs/context_numbers_debug.yaml
+++ b/configs/context_numbers_debug.yaml
@@ -26,7 +26,7 @@ learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
 weight_decay: 0.01
-warmup_ratio: 0.1
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/context_numbers_easy.yaml b/configs/context_numbers_easy.yaml
index 316114a..ff2e56c 100644
--- a/configs/context_numbers_easy.yaml
+++ b/configs/context_numbers_easy.yaml
@@ -26,7 +26,7 @@ learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
 weight_decay: 0.01
-warmup_ratio: 0.1
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/default.yaml b/configs/default.yaml
index 8bf7fa8..bd37401 100644
--- a/configs/default.yaml
+++ b/configs/default.yaml
@@ -24,7 +24,7 @@ learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
 weight_decay: 0.01
-warmup_ratio: 0.1
+
 
 # LoRA
 lora_r: 8
diff --git a/configs/fw_qa.yaml b/configs/fw_qa.yaml
index cc7dcb8..9195a67 100644
--- a/configs/fw_qa.yaml
+++ b/configs/fw_qa.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-warmup_ratio: 0.1
+
 
 dataloader_prefetch_factor: 8
 dataloader_num_workers: 8
diff --git a/configs/fw_qa_ctx_qa.yaml b/configs/fw_qa_ctx_qa.yaml
index 0263f92..941d1e0 100644
--- a/configs/fw_qa_ctx_qa.yaml
+++ b/configs/fw_qa_ctx_qa.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-warmup_ratio: 0.1
+
 
 dataloader_prefetch_factor: 8
 dataloader_num_workers: 8
diff --git a/configs/fw_qa_large_ctx_pwc_hotpot.yaml b/configs/fw_qa_large_ctx_pwc_hotpot.yaml
index 0efd743..4c601b5 100644
--- a/configs/fw_qa_large_ctx_pwc_hotpot.yaml
+++ b/configs/fw_qa_large_ctx_pwc_hotpot.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-warmup_ratio: 0.1
+
 
 dataloader_prefetch_factor: 8
 dataloader_num_workers: 8
diff --git a/configs/fw_qa_large_ctx_pwc_hotpot_squad.yaml b/configs/fw_qa_large_ctx_pwc_hotpot_squad.yaml
index 0c2d194..b4c6e93 100644
--- a/configs/fw_qa_large_ctx_pwc_hotpot_squad.yaml
+++ b/configs/fw_qa_large_ctx_pwc_hotpot_squad.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-# warmup_ratio: 0.1
+# 
 warmup_steps: 100
 
 dataloader_prefetch_factor: 8
diff --git a/configs/fw_qa_pwc_hotpot.yaml b/configs/fw_qa_pwc_hotpot.yaml
index 1188076..117c223 100644
--- a/configs/fw_qa_pwc_hotpot.yaml
+++ b/configs/fw_qa_pwc_hotpot.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-warmup_ratio: 0.1
+
 
 dataloader_prefetch_factor: 8
 dataloader_num_workers: 8
diff --git a/configs/fw_qa_tiny.yaml b/configs/fw_qa_tiny.yaml
index 50598c0..700b845 100644
--- a/configs/fw_qa_tiny.yaml
+++ b/configs/fw_qa_tiny.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-warmup_ratio: 0.1
+
 
 dataloader_prefetch_factor: 8
 dataloader_num_workers: 8
diff --git a/configs/gsm8k.yaml b/configs/gsm8k.yaml
index 06a0840..c25f9b8 100644
--- a/configs/gsm8k.yaml
+++ b/configs/gsm8k.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-# warmup_ratio: 0.1
+# 
 warmup_steps: 100
 
 dataloader_prefetch_factor: 8
diff --git a/configs/hotpot_qa.yaml b/configs/hotpot_qa.yaml
index 1ad924d..10ff35c 100644
--- a/configs/hotpot_qa.yaml
+++ b/configs/hotpot_qa.yaml
@@ -23,8 +23,8 @@ max_val_samples_per_ds: 1000
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.1
+weight_decay: 0.01
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/math_and_code.yaml b/configs/math_and_code.yaml
index 4254ca9..1a9bcf6 100644
--- a/configs/math_and_code.yaml
+++ b/configs/math_and_code.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-# warmup_ratio: 0.1
+# 
 warmup_steps: 100
 
 dataloader_prefetch_factor: 8
diff --git a/configs/openmathintx.yaml b/configs/openmathintx.yaml
index b48092c..eeb5253 100644
--- a/configs/openmathintx.yaml
+++ b/configs/openmathintx.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-# warmup_ratio: 0.1
+# 
 warmup_steps: 100
 
 dataloader_prefetch_factor: 8
diff --git a/configs/pretrain_all.yaml b/configs/pretrain_all.yaml
index 86a22b6..b7926fc 100644
--- a/configs/pretrain_all.yaml
+++ b/configs/pretrain_all.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 5
 weight_decay: 0.01
-# warmup_ratio: 0.1
+# 
 warmup_steps: 100
 
 dataloader_prefetch_factor: 8
diff --git a/configs/pwc.yaml b/configs/pwc.yaml
index 9c7da4d..7344e33 100644
--- a/configs/pwc.yaml
+++ b/configs/pwc.yaml
@@ -23,8 +23,8 @@ max_val_samples_per_ds: 1000
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.1
+weight_decay: 0.01
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/pwc_and_ctx_numbers_256.yaml b/configs/pwc_and_ctx_numbers_256.yaml
index 9c95e68..080ff02 100644
--- a/configs/pwc_and_ctx_numbers_256.yaml
+++ b/configs/pwc_and_ctx_numbers_256.yaml
@@ -23,8 +23,8 @@ max_val_samples_per_ds: 1000
 learning_rate: 0.00001
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
-weight_decay: 0.1
-warmup_ratio: 0.1
+weight_decay: 0.01
+
 
 # LoRA
 lora_r: 16
diff --git a/configs/pwc_hotpot_qa.yaml b/configs/pwc_hotpot_qa.yaml
index e59e494..d2cb994 100644
--- a/configs/pwc_hotpot_qa.yaml
+++ b/configs/pwc_hotpot_qa.yaml
@@ -25,7 +25,7 @@ learning_rate: 0.00002
 # lr_scheduler_type: "constant_with_warmup"
 neftune_noise_alpha: 1
 weight_decay: 0.01
-warmup_ratio: 0.1
+
 
 # LoRA
 lora_r: 8
diff --git a/src/ctx_to_lora/configs.py b/src/ctx_to_lora/configs.py
index d153a6c..91d182f 100644
--- a/src/ctx_to_lora/configs.py
+++ b/src/ctx_to_lora/configs.py
@@ -49,6 +49,11 @@ class ArgumentParser(HfArgumentParser):
             for arg, val in other_args.items():
                 # add only if in keys
                 if arg in keys:
+                    if val in ["None", "none", "null", "NULL"]:
+                        val = None
+                        inputs[arg] = val
+                        used_args[arg] = val
+                        continue
                     base_type = data_yaml.__dataclass_fields__[arg].type
                     inputs[arg] = val
 
@@ -166,6 +171,10 @@ class TrainingArguments(TrainingArguments):
         default=None,
         metadata={"help": "Learning rate scheduler kwargs."},
     )
+    warmup_steps: int = field(
+        default=100,
+        metadata={"help": "Number of warmup steps."},
+    )
     eval_on_start: bool = field(
         default=True,
         metadata={"help": "Whether to evaluate on the start of training."},