From 3907dcb902439c4631d3008791f07d055301783a Mon Sep 17 00:00:00 2001 From: yuiseki Date: Fri, 29 Mar 2024 14:00:01 +0900 Subject: [PATCH] train_claim_gpu_num: 4 --- recipes/A5000_24GB_x8/coder-dolphin-en.yaml | 2 +- recipes/A5000_24GB_x8/coder-math-ja-wikipedia.yaml | 2 +- recipes/A5000_24GB_x8/coder-math-ja.yaml | 2 +- recipes/A5000_24GB_x8/coder-wizardlm-en.yaml | 2 +- 4 files changed, 4 insertions(+), 4 deletions(-) diff --git a/recipes/A5000_24GB_x8/coder-dolphin-en.yaml b/recipes/A5000_24GB_x8/coder-dolphin-en.yaml index de553fe..cf5a702 100644 --- 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 index 5fbb01d..16c8d3a 100644 --- 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 index e4c6a80..446848a 100644 --- 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 index bc14564..747eb2d 100644 --- 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