train_claim_gpu_num: 4

yuiseki · yuiseki · commit 3907dcb90243 · 2024-03-29T14:00:01.000+09:00
diff --git a/recipes/A5000_24GB_x8/coder-dolphin-en.yaml b/recipes/A5000_24GB_x8/coder-dolphin-en.yaml
@@ -12,7 +12,7 @@ dataset_train_split_test_size: 0.2
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.05
-train_claim_gpu_num: 3
+train_claim_gpu_num: 4
 train_per_device_train_batch_size: 8
 train_gradient_accumulation_steps: 2
 train_num_train_epochs: 4
diff --git a/recipes/A5000_24GB_x8/coder-math-ja-wikipedia.yaml b/recipes/A5000_24GB_x8/coder-math-ja-wikipedia.yaml
@@ -10,7 +10,7 @@ dataset_train_split_test_size: 0.2
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.05
-train_claim_gpu_num: 3
+train_claim_gpu_num: 4
 train_per_device_train_batch_size: 8
 train_gradient_accumulation_steps: 2
 train_num_train_epochs: 4
diff --git a/recipes/A5000_24GB_x8/coder-math-ja.yaml b/recipes/A5000_24GB_x8/coder-math-ja.yaml
@@ -10,7 +10,7 @@ dataset_train_split_test_size: 0.2
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.05
-train_claim_gpu_num: 3
+train_claim_gpu_num: 4
 train_per_device_train_batch_size: 8
 train_gradient_accumulation_steps: 2
 train_num_train_epochs: 4
diff --git a/recipes/A5000_24GB_x8/coder-wizardlm-en.yaml b/recipes/A5000_24GB_x8/coder-wizardlm-en.yaml
@@ -10,7 +10,7 @@ dataset_train_split_test_size: 0.2
 lora_r: 8
 lora_alpha: 16
 lora_dropout: 0.05
-train_claim_gpu_num: 3
+train_claim_gpu_num: 4
 train_per_device_train_batch_size: 8
 train_gradient_accumulation_steps: 2
 train_num_train_epochs: 4
