Update dpo_llama3.py

marcopoli · May 11, 2024 · ffbb396 · ffbb396
1 parent eaf49a6
commit ffbb396
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/model_adaptation/dpo_llama3.py b/model_adaptation/dpo_llama3.py
@@ -85,8 +85,8 @@ def apply_dpo_template(example):
         per_device_train_batch_size = 8,
         gradient_accumulation_steps = 12,
         warmup_ratio = 0.1,
-        num_train_epochs = 2,
-        learning_rate = 5e-6,
+        num_train_epochs = 1,
+        learning_rate = 5e-5,
         fp16 = not torch.cuda.is_bf16_supported(),
         bf16 = torch.cuda.is_bf16_supported(),
         logging_steps = 1,