diff --git a/nemo/collections/llm/api.py b/nemo/collections/llm/api.py
index aaef714ef738..4bafdd97ba21 100644
--- a/nemo/collections/llm/api.py
+++ b/nemo/collections/llm/api.py
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
+import json
 import os
 from copy import deepcopy
 from pathlib import Path
diff --git a/nemo/collections/llm/recipes/baichuan2_7b.py b/nemo/collections/llm/recipes/baichuan2_7b.py
index 823f6e07cd57..1350cbaa7edd 100644
--- a/nemo/collections/llm/recipes/baichuan2_7b.py
+++ b/nemo/collections/llm/recipes/baichuan2_7b.py
@@ -25,7 +25,7 @@
 from nemo.collections.llm import Baichuan2Config7B, Baichuan2Model
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -254,8 +254,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -279,8 +281,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/chatglm3_6b.py b/nemo/collections/llm/recipes/chatglm3_6b.py
index b6c640372074..2cd424ce5bf6 100644
--- a/nemo/collections/llm/recipes/chatglm3_6b.py
+++ b/nemo/collections/llm/recipes/chatglm3_6b.py
@@ -25,7 +25,7 @@
 from nemo.collections.llm import ChatGLM3Config6B, ChatGLMModel
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -254,8 +254,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -279,8 +281,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/finetune_default.py b/nemo/collections/llm/recipes/finetune_default.py
index f05fd7cb2d13..e8af7f67bdbd 100644
--- a/nemo/collections/llm/recipes/finetune_default.py
+++ b/nemo/collections/llm/recipes/finetune_default.py
@@ -21,9 +21,11 @@
 import nemo.lightning as nl
 from nemo.collections import llm
 from nemo.collections.llm.gpt.data.packed_sequence import PackedSequenceSpecs
+from nemo.collections.llm.peft import DoRA, LoRA
 from nemo.collections.llm.recipes.log.default import tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
 from nemo.collections.llm.recipes.precision.mixed_precision import bf16_mixed
+from nemo.lightning.pytorch.callbacks import PEFT
 
 
 def default_finetune_recipe(
@@ -158,3 +160,41 @@ def nemo_resume(model_id: str) -> run.Config[nl.AutoResume]:
         nl.AutoResume,
         restore_config=run.Config(nl.RestoreConfig, path=f"nemo://{model_id}"),
     )
+
+
+@run.cli.factory(name='lora')
+def lora() -> run.Config[PEFT]:
+    """
+    Factory function to create a LoRA configuration.
+
+    Returns:
+        run.Config[PEFT]: Configuration for the LoRA class.
+
+    Examples:
+        CLI usage:
+            $ nemo llm finetune -f llama3_8b peft=lora
+
+        Python API usage:
+            >>> lora_config = lora()
+            >>> print(lora_config)
+    """
+    return run.Config(LoRA)
+
+
+@run.cli.factory(name='dora')
+def dora() -> run.Config[PEFT]:
+    """
+    Factory function to create a DoRA configuration.
+
+    Returns:
+        run.Config[PEFT]: Configuration for the DoRA class.
+
+    Examples:
+        CLI usage:
+            $ nemo llm finetune -f llama3_8b peft=dora
+
+        Python API usage:
+            >>> dora_config = dora()
+            >>> print(dora_config)
+    """
+    return run.Config(DoRA)
diff --git a/nemo/collections/llm/recipes/gemma2_27b.py b/nemo/collections/llm/recipes/gemma2_27b.py
index 2025bd570503..d6b41c0a221c 100644
--- a/nemo/collections/llm/recipes/gemma2_27b.py
+++ b/nemo/collections/llm/recipes/gemma2_27b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.gemma2 import gemma2_model, gemma2_trainer
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
@@ -191,8 +191,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -220,8 +222,8 @@ def finetune_recipe(
         recipe.optim.config.lr = 5e-6
         recipe.trainer.strategy.tensor_model_parallel_size = 8
         recipe.trainer.strategy.pipeline_model_parallel_size = 2
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.trainer.strategy.tensor_model_parallel_size = 4
         recipe.optim.config.lr = 1e-4
     else:
diff --git a/nemo/collections/llm/recipes/gemma2_2b.py b/nemo/collections/llm/recipes/gemma2_2b.py
index e1aa3ad4be86..138140d0515d 100644
--- a/nemo/collections/llm/recipes/gemma2_2b.py
+++ b/nemo/collections/llm/recipes/gemma2_2b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.gemma2 import gemma2_model, gemma2_trainer
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
@@ -191,8 +191,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -218,8 +220,8 @@ def finetune_recipe(
 
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/gemma2_9b.py b/nemo/collections/llm/recipes/gemma2_9b.py
index 8117102f1b75..c49ac0246307 100644
--- a/nemo/collections/llm/recipes/gemma2_9b.py
+++ b/nemo/collections/llm/recipes/gemma2_9b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.gemma2 import gemma2_model, gemma2_trainer
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
@@ -191,8 +191,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -219,8 +221,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.optim.config.lr = 5e-6
         recipe.trainer.strategy.tensor_model_parallel_size = 4
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/gemma_2b.py b/nemo/collections/llm/recipes/gemma_2b.py
index 8798af436a9c..8bdf89696d56 100644
--- a/nemo/collections/llm/recipes/gemma_2b.py
+++ b/nemo/collections/llm/recipes/gemma_2b.py
@@ -24,7 +24,7 @@
 from nemo.collections.llm import GemmaConfig2B, GemmaModel
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -253,8 +253,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given maximum seq_length for better efficiency.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -284,8 +286,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.context_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/gemma_7b.py b/nemo/collections/llm/recipes/gemma_7b.py
index 0bfd62b33e9e..46c91e27575a 100644
--- a/nemo/collections/llm/recipes/gemma_7b.py
+++ b/nemo/collections/llm/recipes/gemma_7b.py
@@ -24,7 +24,7 @@
 from nemo.collections.llm import GemmaConfig7B, GemmaModel
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -256,8 +256,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -287,8 +289,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/llama31_405b.py b/nemo/collections/llm/recipes/llama31_405b.py
index d71f3791a0af..5f08d82bd888 100644
--- a/nemo/collections/llm/recipes/llama31_405b.py
+++ b/nemo/collections/llm/recipes/llama31_405b.py
@@ -26,7 +26,7 @@
 from nemo.collections.llm.gpt.data.mock import MockDataModule
 from nemo.collections.llm.gpt.data.packed_sequence import PackedSequenceSpecs
 from nemo.collections.llm.gpt.model.llama import Llama31Config405B, LlamaModel
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -266,7 +266,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for finetuning. Allowed values: 'lora'/'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
         seq_length (int): Maximum number of tokens per microbatch.
         packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given
             maximum seq_length for better efficiency. By default, this value equals performance_mode.
@@ -296,7 +297,7 @@ def finetune_recipe(
     if num_nodes is None:
         if peft_scheme is None or peft_scheme.lower() == 'none':
             num_nodes = 12
-        elif peft_scheme.lower() == 'lora':
+        elif peft_scheme.lower() in ['lora', 'dora']:
             num_nodes = 3
 
     recipe = default_finetune_recipe(
@@ -307,8 +308,8 @@ def finetune_recipe(
         recipe.trainer.strategy.pipeline_model_parallel_size = 14
         recipe.data.global_batch_size = 6
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.peft.dim = 16
         recipe.peft.alpha = 32
         recipe.optim.config.use_distributed_optimizer = False
@@ -348,7 +349,8 @@ def finetune_performance_optimizations(
 
     Args:
         recipe (run.Partial): Base fine-tuning recipe to which performance optimizations will be added
-        peft_scheme (str): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
 
     Returns:
         run.Partial: Partial configuration for performance-optimized fine-tuning.
diff --git a/nemo/collections/llm/recipes/llama31_70b.py b/nemo/collections/llm/recipes/llama31_70b.py
index 37e809e5bc8f..3120fedd7923 100644
--- a/nemo/collections/llm/recipes/llama31_70b.py
+++ b/nemo/collections/llm/recipes/llama31_70b.py
@@ -26,7 +26,7 @@
 from nemo.collections.llm.gpt.data.mock import MockDataModule
 from nemo.collections.llm.gpt.data.packed_sequence import PackedSequenceSpecs
 from nemo.collections.llm.gpt.model.llama import Llama31Config70B, LlamaModel
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -266,7 +266,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for finetuning. Allowed values: 'lora'/'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
         seq_length (int): Maximum number of tokens per microbatch.
         packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given
             maximum seq_length for better efficiency. By default, this value equals performance_mode.
@@ -300,7 +301,7 @@ def finetune_recipe(
     if num_nodes is None:
         if peft_scheme is None or peft_scheme.lower() == 'none':
             num_nodes = 4
-        elif peft_scheme.lower() == 'lora':
+        elif peft_scheme.lower() in ['lora', 'dora']:
             num_nodes = 1
 
     recipe = default_finetune_recipe(
@@ -310,8 +311,8 @@ def finetune_recipe(
         recipe.trainer.strategy.tensor_model_parallel_size = 8
         recipe.trainer.strategy.pipeline_model_parallel_size = 4
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.peft.dim = 16
         recipe.peft.alpha = 32
         recipe.optim.config.use_distributed_optimizer = False
@@ -349,7 +350,8 @@ def finetune_performance_optimizations(
 
     Args:
         recipe (run.Partial): Base fine-tuning recipe to which performance optimizations will be added
-        peft_scheme (str): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
 
     Returns:
         run.Partial: Partial configuration for performance-optimized fine-tuning.
diff --git a/nemo/collections/llm/recipes/llama31_8b.py b/nemo/collections/llm/recipes/llama31_8b.py
index 32a77ce076f2..62514940b678 100644
--- a/nemo/collections/llm/recipes/llama31_8b.py
+++ b/nemo/collections/llm/recipes/llama31_8b.py
@@ -26,7 +26,7 @@
 from nemo.collections.llm.gpt.data.mock import MockDataModule
 from nemo.collections.llm.gpt.data.packed_sequence import PackedSequenceSpecs
 from nemo.collections.llm.gpt.model.llama import Llama31Config8B, LlamaModel
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -266,7 +266,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for finetuning. Allowed values: 'lora'/'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
         seq_length (int): Maximum number of tokens per microbatch.
         packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given
             maximum seq_length for better efficiency. By default, this value equals performance_mode.
@@ -303,8 +304,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.peft.dim = 8
         recipe.peft.alpha = 16
         recipe.optim.config.use_distributed_optimizer = False
@@ -341,7 +342,8 @@ def finetune_performance_optimizations(
 
     Args:
         recipe (run.Partial): Base fine-tuning recipe to which performance optimizations will be added
-        peft_scheme (str): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
 
     Returns:
         run.Partial: Partial configuration for performance-optimized fine-tuning.
diff --git a/nemo/collections/llm/recipes/llama3_70b.py b/nemo/collections/llm/recipes/llama3_70b.py
index 93aeb5c07dc1..8b61bff80e01 100644
--- a/nemo/collections/llm/recipes/llama3_70b.py
+++ b/nemo/collections/llm/recipes/llama3_70b.py
@@ -26,7 +26,7 @@
 from nemo.collections.llm.gpt.data.mock import MockDataModule
 from nemo.collections.llm.gpt.data.packed_sequence import PackedSequenceSpecs
 from nemo.collections.llm.gpt.model.llama import Llama3Config70B, LlamaModel
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -263,7 +263,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for finetuning. Allowed values: 'lora'/'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
         seq_length (int): Maximum number of tokens per microbatch.
         packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given
             maximum seq_length for better efficiency. By default, this value equals performance_mode.
@@ -297,7 +298,7 @@ def finetune_recipe(
     if num_nodes is None:
         if peft_scheme is None or peft_scheme.lower() == 'none':
             num_nodes = 4
-        elif peft_scheme.lower() == 'lora':
+        elif peft_scheme.lower() in ['lora', 'dora']:
             num_nodes = 1
 
     recipe = default_finetune_recipe(
@@ -307,8 +308,8 @@ def finetune_recipe(
         recipe.trainer.strategy.tensor_model_parallel_size = 8
         recipe.trainer.strategy.pipeline_model_parallel_size = 4
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.peft.dim = 16
         recipe.peft.alpha = 32
         recipe.optim.config.use_distributed_optimizer = False
@@ -346,7 +347,8 @@ def finetune_performance_optimizations(
 
     Args:
         recipe (run.Partial): Base fine-tuning recipe to which performance optimizations will be added
-        peft_scheme (str): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
 
     Returns:
         run.Partial: Partial configuration for performance-optimized fine-tuning.
diff --git a/nemo/collections/llm/recipes/llama3_8b.py b/nemo/collections/llm/recipes/llama3_8b.py
index 3e26f8fe1082..36b20c12ddb2 100644
--- a/nemo/collections/llm/recipes/llama3_8b.py
+++ b/nemo/collections/llm/recipes/llama3_8b.py
@@ -26,7 +26,7 @@
 from nemo.collections.llm.gpt.data.mock import MockDataModule
 from nemo.collections.llm.gpt.data.packed_sequence import PackedSequenceSpecs
 from nemo.collections.llm.gpt.model.llama import Llama3Config8B, LlamaModel
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -250,7 +250,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for finetuning. Allowed values: 'lora'/'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
         seq_length (int): Maximum number of tokens per microbatch.
         packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given
             maximum seq_length for better efficiency. By default, this value equals performance_mode.
@@ -287,8 +288,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.peft.dim = 8
         recipe.peft.alpha = 16
         recipe.optim.config.use_distributed_optimizer = False
@@ -325,7 +326,8 @@ def finetune_performance_optimizations(
 
     Args:
         recipe (run.Partial): Base fine-tuning recipe to which performance optimizations will be added
-        peft_scheme (str): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
 
     Returns:
         run.Partial: Partial configuration for performance-optimized fine-tuning.
diff --git a/nemo/collections/llm/recipes/mistral_7b.py b/nemo/collections/llm/recipes/mistral_7b.py
index 3bc1e568185a..9e2d2e256fbe 100644
--- a/nemo/collections/llm/recipes/mistral_7b.py
+++ b/nemo/collections/llm/recipes/mistral_7b.py
@@ -24,9 +24,8 @@
 from nemo import lightning as nl
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.gpt.data.squad import SquadDataModule
 from nemo.collections.llm.gpt.model.mistral import MistralConfig7B, MistralModel
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -207,8 +206,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -237,8 +238,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/mistral_nemo_12b.py b/nemo/collections/llm/recipes/mistral_nemo_12b.py
index 7d9fa1d792e9..a10f8ae804b8 100644
--- a/nemo/collections/llm/recipes/mistral_nemo_12b.py
+++ b/nemo/collections/llm/recipes/mistral_nemo_12b.py
@@ -24,9 +24,8 @@
 from nemo import lightning as nl
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.gpt.data.squad import SquadDataModule
 from nemo.collections.llm.gpt.model.mistral import MistralModel, MistralNeMoConfig12B
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -255,8 +254,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -285,8 +286,10 @@ def finetune_recipe(
     )
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA, target_modules=['linear_qkv', 'linear_proj'], dim=32)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(
+            PEFT_STR2CLS[peft_scheme.lower()], target_modules=['linear_qkv', 'linear_proj'], dim=32
+        )
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/mixtral_8x22b.py b/nemo/collections/llm/recipes/mixtral_8x22b.py
index 16e6168e649b..ec1641a08d80 100644
--- a/nemo/collections/llm/recipes/mixtral_8x22b.py
+++ b/nemo/collections/llm/recipes/mixtral_8x22b.py
@@ -24,9 +24,8 @@
 from nemo import lightning as nl
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.gpt.data.squad import SquadDataModule
 from nemo.collections.llm.gpt.model.mixtral import MixtralConfig8x22B, MixtralModel
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -227,7 +226,7 @@ def pretrain_performance_optimizations(recipe: run.Partial) -> run.Partial:
             ),
             run.Config(
                 MegatronCommOverlapCallback,
-                overlap_param_gather_with_optimizer_step=False,  # Currently disabled due to an issue with checkpointing
+                overlap_param_gather_with_optimizer_step=False,  # Currently disabled due to issue with checkpointing
                 align_param_gather=True,
             ),
         ]
@@ -259,8 +258,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given maximum seq_length for better efficiency.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given
+            maximum seq_length for better efficiency.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -286,8 +287,10 @@ def finetune_recipe(
         recipe.trainer.strategy.pipeline_model_parallel_size = 4
         recipe.trainer.strategy.virtual_pipeline_model_parallel_size = 14
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA, target_modules=['linear_qkv', 'linear_proj'], dim=32)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(
+            PEFT_STR2CLS[peft_scheme.lower()], target_modules=['linear_qkv', 'linear_proj'], dim=32
+        )
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/mixtral_8x7b.py b/nemo/collections/llm/recipes/mixtral_8x7b.py
index 5fbb0ac22c61..d06e22fc2180 100644
--- a/nemo/collections/llm/recipes/mixtral_8x7b.py
+++ b/nemo/collections/llm/recipes/mixtral_8x7b.py
@@ -24,9 +24,8 @@
 from nemo import lightning as nl
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.gpt.data.squad import SquadDataModule
 from nemo.collections.llm.gpt.model.mixtral import MixtralConfig8x7B, MixtralModel
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -222,7 +221,7 @@ def pretrain_performance_optimizations(recipe: run.Partial) -> run.Partial:
             run.Config(MegatronTokenDropCallback),
             run.Config(
                 MegatronCommOverlapCallback,
-                overlap_param_gather_with_optimizer_step=False,  # Currently disabled due to an issue with checkpointing.
+                overlap_param_gather_with_optimizer_step=False,  # Currently disabled due to issue with checkpointing.
                 align_param_gather=True,
             ),
         ]
@@ -254,8 +253,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -280,8 +281,10 @@ def finetune_recipe(
         recipe.trainer.strategy.pipeline_model_parallel_size = 4
         recipe.trainer.strategy.virtual_pipeline_model_parallel_size = 8
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA, target_modules=['linear_qkv', 'linear_proj'], dim=32)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(
+            PEFT_STR2CLS[peft_scheme.lower()], target_modules=['linear_qkv', 'linear_proj'], dim=32
+        )
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/nemotron3_22b.py b/nemo/collections/llm/recipes/nemotron3_22b.py
index 2dd9c3ff5205..4c763301bc52 100644
--- a/nemo/collections/llm/recipes/nemotron3_22b.py
+++ b/nemo/collections/llm/recipes/nemotron3_22b.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Callable, Optional
+from typing import Optional
 
 import lightning.pytorch as pl
 import nemo_run as run
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.nemotron import nemotron_model, nemotron_trainer
@@ -239,8 +239,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -265,8 +267,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 8
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/nemotron3_4b.py b/nemo/collections/llm/recipes/nemotron3_4b.py
index c208ee740265..fc6f09a09358 100644
--- a/nemo/collections/llm/recipes/nemotron3_4b.py
+++ b/nemo/collections/llm/recipes/nemotron3_4b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.nemotron import nemotron_model, nemotron_trainer
@@ -191,8 +191,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -216,8 +218,8 @@ def finetune_recipe(
     )
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/nemotron3_8b.py b/nemo/collections/llm/recipes/nemotron3_8b.py
index 7799512c6260..f60463330cad 100644
--- a/nemo/collections/llm/recipes/nemotron3_8b.py
+++ b/nemo/collections/llm/recipes/nemotron3_8b.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Callable, Optional
+from typing import Optional
 
 import lightning.pytorch as pl
 import nemo_run as run
@@ -21,8 +21,7 @@
 from nemo import lightning as nl
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.gpt.data.squad import SquadDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.nemotron import nemotron_model, nemotron_trainer
@@ -256,8 +255,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -282,8 +283,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/nemotron4_15b.py b/nemo/collections/llm/recipes/nemotron4_15b.py
index ad0f884b0d3b..49f92fcc1616 100644
--- a/nemo/collections/llm/recipes/nemotron4_15b.py
+++ b/nemo/collections/llm/recipes/nemotron4_15b.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Callable, Optional
+from typing import Optional
 
 import lightning.pytorch as pl
 import nemo_run as run
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.nemotron import nemotron_model, nemotron_trainer
@@ -228,8 +228,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -254,8 +256,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 4
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/nemotron4_340b.py b/nemo/collections/llm/recipes/nemotron4_340b.py
index b22abc43d558..14d4c0f32d11 100644
--- a/nemo/collections/llm/recipes/nemotron4_340b.py
+++ b/nemo/collections/llm/recipes/nemotron4_340b.py
@@ -12,16 +12,15 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Callable, Optional
+from typing import Optional
 
 import lightning.pytorch as pl
 import nemo_run as run
 import torch
 
-from nemo import lightning as nl
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.nemotron import nemotron_model, nemotron_trainer
@@ -240,8 +239,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -268,8 +269,8 @@ def finetune_recipe(
         recipe.trainer.strategy.tensor_model_parallel_size = 8
         recipe.trainer.strategy.pipeline_model_parallel_size = 12
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.trainer.strategy.tensor_model_parallel_size = 8
         recipe.trainer.strategy.pipeline_model_parallel_size = 4
         recipe.optim.config.lr = 1e-4
diff --git a/nemo/collections/llm/recipes/phi3_mini_4k_instruct.py b/nemo/collections/llm/recipes/phi3_mini_4k_instruct.py
index 1cbc877dc33e..73bbe4735adb 100644
--- a/nemo/collections/llm/recipes/phi3_mini_4k_instruct.py
+++ b/nemo/collections/llm/recipes/phi3_mini_4k_instruct.py
@@ -25,7 +25,7 @@
 from nemo.collections.llm.gpt.data.mock import MockDataModule
 from nemo.collections.llm.gpt.data.packed_sequence import PackedSequenceSpecs
 from nemo.collections.llm.gpt.model.phi3mini import Phi3ConfigMini, Phi3Model
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -222,8 +222,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora',
-        'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
         seq_length (int): Maximum number of tokens per microbatch.
         packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given
         maximum seq_length for better efficiency. By default, this value equals performance_mode.
@@ -260,8 +260,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 1
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.peft.dim = 8
         recipe.peft.alpha = 16
         recipe.optim.config.use_distributed_optimizer = False
diff --git a/nemo/collections/llm/recipes/qwen2_1p5b.py b/nemo/collections/llm/recipes/qwen2_1p5b.py
index a3d705c4fb3a..99ba5cd907fc 100644
--- a/nemo/collections/llm/recipes/qwen2_1p5b.py
+++ b/nemo/collections/llm/recipes/qwen2_1p5b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -194,8 +194,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -218,8 +220,8 @@ def finetune_recipe(
     )
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/qwen2_500m.py b/nemo/collections/llm/recipes/qwen2_500m.py
index 08541ca9e421..96d99c271c85 100644
--- a/nemo/collections/llm/recipes/qwen2_500m.py
+++ b/nemo/collections/llm/recipes/qwen2_500m.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -194,8 +194,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -218,8 +220,8 @@ def finetune_recipe(
     )
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/qwen2_72b.py b/nemo/collections/llm/recipes/qwen2_72b.py
index c0bc9bf40611..33bb0dd40835 100644
--- a/nemo/collections/llm/recipes/qwen2_72b.py
+++ b/nemo/collections/llm/recipes/qwen2_72b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -194,8 +194,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -221,8 +223,8 @@ def finetune_recipe(
         recipe.trainer.strategy.tensor_model_parallel_size = 8
         recipe.trainer.strategy.pipeline_model_parallel_size = 4
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.trainer.strategy.tensor_model_parallel_size = 8
         recipe.optim.config.lr = 1e-4
     else:
diff --git a/nemo/collections/llm/recipes/qwen2_7b.py b/nemo/collections/llm/recipes/qwen2_7b.py
index 67bcc5e953bf..2e62176a408e 100644
--- a/nemo/collections/llm/recipes/qwen2_7b.py
+++ b/nemo/collections/llm/recipes/qwen2_7b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -194,8 +194,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -219,8 +221,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/starcoder2_15b.py b/nemo/collections/llm/recipes/starcoder2_15b.py
index 14b53809111a..e424cb67dba4 100644
--- a/nemo/collections/llm/recipes/starcoder2_15b.py
+++ b/nemo/collections/llm/recipes/starcoder2_15b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -194,8 +194,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -219,8 +221,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 4
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/starcoder2_3b.py b/nemo/collections/llm/recipes/starcoder2_3b.py
index 3ee81522ebc9..faf0b416c56a 100644
--- a/nemo/collections/llm/recipes/starcoder2_3b.py
+++ b/nemo/collections/llm/recipes/starcoder2_3b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -194,8 +194,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -219,8 +221,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/starcoder2_7b.py b/nemo/collections/llm/recipes/starcoder2_7b.py
index 96b5ab36b876..091e882cd932 100644
--- a/nemo/collections/llm/recipes/starcoder2_7b.py
+++ b/nemo/collections/llm/recipes/starcoder2_7b.py
@@ -20,7 +20,7 @@
 
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -194,8 +194,10 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
-        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training efficiency. Default sequence length is 2048.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
+        packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training
+            efficiency. Default sequence length is 2048.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -219,8 +221,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/starcoder_15b.py b/nemo/collections/llm/recipes/starcoder_15b.py
index d87788be5613..382d0eb4d8ca 100644
--- a/nemo/collections/llm/recipes/starcoder_15b.py
+++ b/nemo/collections/llm/recipes/starcoder_15b.py
@@ -23,7 +23,7 @@
 from nemo.collections.llm.api import finetune, pretrain
 from nemo.collections.llm.gpt.data.mock import MockDataModule
 from nemo.collections.llm.gpt.model.starcoder import StarcoderConfig15B, StarcoderModel
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_recipe
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing
@@ -280,7 +280,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -302,8 +303,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.pipeline_model_parallel_size = 8
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/t5_11b.py b/nemo/collections/llm/recipes/t5_11b.py
index 8baf54b4f42f..ee7323aa044f 100644
--- a/nemo/collections/llm/recipes/t5_11b.py
+++ b/nemo/collections/llm/recipes/t5_11b.py
@@ -24,7 +24,7 @@
 
 from nemo import lightning as nl
 from nemo.collections.llm.api import finetune, pretrain
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_trainer, nemo_resume
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.precision.mixed_precision import bf16_mixed
@@ -229,7 +229,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -279,8 +280,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 4
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/t5_220m.py b/nemo/collections/llm/recipes/t5_220m.py
index 27feb43837fb..edc9fdba62d7 100644
--- a/nemo/collections/llm/recipes/t5_220m.py
+++ b/nemo/collections/llm/recipes/t5_220m.py
@@ -24,7 +24,7 @@
 
 from nemo import lightning as nl
 from nemo.collections.llm.api import finetune, pretrain
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_trainer, nemo_resume
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.precision.mixed_precision import bf16_mixed
@@ -229,7 +229,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -279,8 +280,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 1
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")
diff --git a/nemo/collections/llm/recipes/t5_3b.py b/nemo/collections/llm/recipes/t5_3b.py
index 333661d97117..82772e1b865a 100644
--- a/nemo/collections/llm/recipes/t5_3b.py
+++ b/nemo/collections/llm/recipes/t5_3b.py
@@ -24,7 +24,7 @@
 
 from nemo import lightning as nl
 from nemo.collections.llm.api import finetune, pretrain
-from nemo.collections.llm.peft.lora import LoRA
+from nemo.collections.llm.peft import PEFT_STR2CLS
 from nemo.collections.llm.recipes.finetune_default import default_finetune_trainer, nemo_resume
 from nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger
 from nemo.collections.llm.recipes.precision.mixed_precision import bf16_mixed
@@ -229,7 +229,8 @@ def finetune_recipe(
         name (str): Name of the fine-tuning run.
         num_nodes (int): Number of compute nodes to use.
         num_gpus_per_node (int): Number of GPUs per node.
-        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora', 'none'/None.
+        peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.
+            Allowed values: 'lora'/'dora'/'none'/None.
 
     Returns:
         run.Partial: Partial configuration for fine-tuning.
@@ -279,8 +280,8 @@ def finetune_recipe(
     if peft_scheme is None or peft_scheme.lower() == 'none':
         recipe.trainer.strategy.tensor_model_parallel_size = 2
         recipe.optim.config.lr = 5e-6
-    elif peft_scheme.lower() == 'lora':
-        recipe.peft = run.Config(LoRA)
+    elif peft_scheme.lower() in ['lora', 'dora']:
+        recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])
         recipe.optim.config.lr = 1e-4
     else:
         raise ValueError(f"Unrecognized peft scheme: {peft_scheme}")