update recipes for 9b and 27b

pytorch · Oct 24, 2024 · 0d53660 · 0d53660
1 parent 6f89920
commit 0d53660
Show file tree

Hide file tree

Showing 8 changed files with 25 additions and 25 deletions.
diff --git a/recipes/configs/gemma2/27B_full.yaml b/recipes/configs/gemma2/27B_full.yaml
@@ -19,7 +19,7 @@
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
-  path: /tmp/gemma2-27b/tokenizer.model
+  path: /tmp/gemma-2-27b/tokenizer.model
 
 # Dataset
 dataset:
@@ -33,12 +33,12 @@ model:
 
 checkpointer:
   _component_: torchtune.training.FullModelHFCheckpointer
-  checkpoint_dir: /tmp/gemma2-27b/
+  checkpoint_dir: /tmp/gemma-2-27b/
   checkpoint_files:
     filename_format: model-{}-of-{}.safetensors
     max_filename: 00024
   recipe_checkpoint: null
-  output_dir: /tmp/gemma2-27b
+  output_dir: /tmp/gemma-2-27b
   model_type: GEMMA2
 resume_from_checkpoint: False
 

diff --git a/recipes/configs/gemma2/27B_lora.yaml b/recipes/configs/gemma2/27B_lora.yaml
@@ -19,7 +19,7 @@
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
-  path: /tmp/gemma2-27b/tokenizer.model
+  path: /tmp/gemma-2-27b/tokenizer.model
 
 # Dataset
 dataset:
@@ -38,12 +38,12 @@ model:
 
 checkpointer:
   _component_: torchtune.training.FullModelHFCheckpointer
-  checkpoint_dir: /tmp/gemma2-27b/
+  checkpoint_dir: /tmp/gemma-2-27b/
   checkpoint_files:
     filename_format: model-{}-of-{}.safetensors
     max_filename: 00024
   recipe_checkpoint: null
-  output_dir: /tmp/gemma2-27b/
+  output_dir: /tmp/gemma-2-27b/
   model_type: GEMMA2
 resume_from_checkpoint: False
 save_adapter_weights_only: False

diff --git a/recipes/configs/gemma2/27B_lora_single_device.yaml b/recipes/configs/gemma2/27B_lora_single_device.yaml
@@ -18,7 +18,7 @@
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
-  path: /tmp/gemma2-27b/tokenizer.model
+  path: /tmp/gemma-2-27b/tokenizer.model
 
 # Dataset
 dataset:
@@ -37,12 +37,12 @@ model:
 
 checkpointer:
   _component_: torchtune.training.FullModelHFCheckpointer
-  checkpoint_dir: /tmp/gemma2-27b/
+  checkpoint_dir: /tmp/gemma-2-27b/
   checkpoint_files:
     filename_format: model-{}-of-{}.safetensors
     max_filename: 00024
   recipe_checkpoint: null
-  output_dir: /tmp/gemma2-27b/
+  output_dir: /tmp/gemma-2-27b/
   model_type: GEMMA2
 resume_from_checkpoint: False
 save_adapter_weights_only: False

diff --git a/recipes/configs/gemma2/27B_qlora_single_device.yaml b/recipes/configs/gemma2/27B_qlora_single_device.yaml
@@ -18,7 +18,7 @@
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
-  path: /tmp/gemma2-27b/tokenizer.model
+  path: /tmp/gemma-2-27b/tokenizer.model
 
 # Dataset
 dataset:
@@ -37,12 +37,12 @@ model:
 
 checkpointer:
   _component_: torchtune.training.FullModelHFCheckpointer
-  checkpoint_dir: /tmp/gemma2-27b/
+  checkpoint_dir: /tmp/gemma-2-27b/
   checkpoint_files:
     filename_format: model-{}-of-{}.safetensors
     max_filename: 00024
   recipe_checkpoint: null
-  output_dir: /tmp/gemma2-27b/
+  output_dir: /tmp/gemma-2-27b/
   model_type: GEMMA2
 resume_from_checkpoint: False
 save_adapter_weights_only: False

diff --git a/recipes/configs/gemma2/9B_full.yaml b/recipes/configs/gemma2/9B_full.yaml
@@ -19,7 +19,7 @@
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
-  path: /tmp/gemma2-9b/tokenizer.model
+  path: /tmp/gemma-2-9b/tokenizer.model
 
 # Dataset
 dataset:
@@ -33,12 +33,12 @@ model:
 
 checkpointer:
   _component_: torchtune.training.FullModelHFCheckpointer
-  checkpoint_dir: /tmp/gemma2-9b/
+  checkpoint_dir: /tmp/gemma-2-9b/
   checkpoint_files:
     filename_format: model-{}-of-{}.safetensors
     max_filename: 00008
   recipe_checkpoint: null
-  output_dir: /tmp/gemma2-9b
+  output_dir: /tmp/gemma-2-9b
   model_type: GEMMA2
 resume_from_checkpoint: False
 

diff --git a/recipes/configs/gemma2/9B_lora.yaml b/recipes/configs/gemma2/9B_lora.yaml
@@ -19,7 +19,7 @@
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
-  path: /tmp/gemma2-9b/tokenizer.model
+  path: /tmp/gemma-2-9b/tokenizer.model
 
 # Dataset
 dataset:
@@ -38,12 +38,12 @@ model:
 
 checkpointer:
   _component_: torchtune.training.FullModelHFCheckpointer
-  checkpoint_dir: /tmp/gemma2-9b/
+  checkpoint_dir: /tmp/gemma-2-9b/
   checkpoint_files:
     filename_format: model-{}-of-{}.safetensors
     max_filename: 00008
   recipe_checkpoint: null
-  output_dir: /tmp/gemma2-9b/
+  output_dir: /tmp/gemma-2-9b/
   model_type: GEMMA2
 resume_from_checkpoint: False
 save_adapter_weights_only: False

diff --git a/recipes/configs/gemma2/9B_lora_single_device.yaml b/recipes/configs/gemma2/9B_lora_single_device.yaml
@@ -18,7 +18,7 @@
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
-  path: /tmp/gemma2-9b/tokenizer.model
+  path: /tmp/gemma-2-9b/tokenizer.model
 
 # Dataset
 dataset:
@@ -37,12 +37,12 @@ model:
 
 checkpointer:
   _component_: torchtune.training.FullModelHFCheckpointer
-  checkpoint_dir: /tmp/gemma2-9b/
+  checkpoint_dir: /tmp/gemma-2-9b/
   checkpoint_files:
     filename_format: model-{}-of-{}.safetensors
     max_filename: 00008
   recipe_checkpoint: null
-  output_dir: /tmp/gemma2-9b/
+  output_dir: /tmp/gemma-2-9b/
   model_type: GEMMA2
 resume_from_checkpoint: False
 save_adapter_weights_only: False

diff --git a/recipes/configs/gemma2/9B_qlora_single_device.yaml b/recipes/configs/gemma2/9B_qlora_single_device.yaml
@@ -18,7 +18,7 @@
 # Tokenizer
 tokenizer:
   _component_: torchtune.models.gemma.gemma_tokenizer
-  path: /tmp/gemma2-9b/tokenizer.model
+  path: /tmp/gemma-2-9b/tokenizer.model
 
 # Dataset
 dataset:
@@ -28,7 +28,7 @@ shuffle: True
 
 # Model Arguments
 model:
-  _component_: torchtune.models.gemma2.qlora_gemma_9b
+  _component_: torchtune.models.gemma2.qlora_gemma2_9b
   lora_attn_modules: ['q_proj', 'k_proj', 'v_proj']
   apply_lora_to_mlp: True
   lora_rank: 64
@@ -37,12 +37,12 @@ model:
 
 checkpointer:
   _component_: torchtune.training.FullModelHFCheckpointer
-  checkpoint_dir: /tmp/gemma2-9b/
+  checkpoint_dir: /tmp/gemma-2-9b/
   checkpoint_files:
     filename_format: model-{}-of-{}.safetensors
     max_filename: 00008
   recipe_checkpoint: null
-  output_dir: /tmp/gemma2-9b/
+  output_dir: /tmp/gemma-2-9b/
   model_type: GEMMA2
 resume_from_checkpoint: False
 save_adapter_weights_only: False