Param Testing Process Image

평가방법

• val , train 비슷한 범위에 속하는지
• val , train 감소하고 있는지
• val set < 0.5

---

---

learning_rates = [5e-5, 3e-5]
batch_sizes = [4]
num_epochs_list = [3]

---

learning_rates = [5e-5, 3e-5]
batch_sizes = [4]
num_epochs_list = [5]

---

learning_rates = [5e-5, 3e-5]
batch_sizes = [8]
num_epochs_list = [3]

---

learning_rates = [5e-5, 3e-5] batch_sizes = [8] num_epochs_list = [5]

---

lr=6e-05, batch_size=2, epochs=7 , linear warmup with decay
🔹 Final Avg Train Loss for lr=6e-05, batch_size=2, epochs=7: 0.5268 🔹 Final Avg Val Loss for lr=6e-05, batch_size=2, epochs=7: 0.5470

---

lr=4e-05, batch_size=2, epochs=8 , linear warmup with decay
🔹 Final Avg Train Loss for lr=4e-05, batch_size=2, epochs=8: 0.5802 🔹 Final Avg Val Loss for lr=4e-05, batch_size=2, epochs=8: 0.5968

---

learning_rates = 7e-05 , batch_sizes = 2 , num_epochs_list = 8 , linear warmup with decay