2025-05-29 nightly release (86f148b)

Author: pytorchbot

recipes/configs/gemma/2B_full.yaml

@@ -23,13 +23,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-2b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma.gemma_2b

recipes/configs/gemma/2B_lora.yaml

@@ -22,13 +22,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-2b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma.lora_gemma_2b

recipes/configs/gemma/7B_full.yaml

@@ -23,13 +23,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-7b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma.gemma_7b

recipes/configs/gemma/7B_lora.yaml

@@ -23,13 +23,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-7b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma.lora_gemma_7b

recipes/configs/gemma2/27B_full.yaml

@@ -23,13 +23,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-2-27b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma2.gemma2_27b

recipes/configs/gemma2/27B_lora.yaml

@@ -23,13 +23,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-2-27b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma2.lora_gemma2_27b

recipes/configs/gemma2/2B_full.yaml

@@ -23,13 +23,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-2-2b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma2.gemma2_2b

recipes/configs/gemma2/2B_lora.yaml

@@ -22,13 +22,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-2-2b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma2.lora_gemma2_2b

recipes/configs/gemma2/9B_full.yaml

@@ -23,13 +23,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-2-9b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma2.gemma2_9b

recipes/configs/gemma2/9B_lora.yaml

@@ -23,13 +23,21 @@ tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
   path: /tmp/gemma-2-9b/tokenizer.model
 
-# Dataset
+# Dataset and Sampler
 dataset:
-  _component_: torchtune.datasets.alpaca_dataset
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
   packed: False  # True increases speed
+  split: train[:95%]
 seed: null
 shuffle: True
 
+# Validation
+run_val_every_n_steps: null  # Change to an integer to enable validation every N steps
+dataset_val:
+  _component_: torchtune.datasets.alpaca_cleaned_dataset
+  split: train[95%:]
+batch_size_val: ${batch_size}
+
 # Model Arguments
 model:
   _component_: torchtune.models.gemma2.lora_gemma2_9b