remove evaluation because it leads to crash due to oom on my pc

train.py

@@ -12,10 +12,10 @@ from transformers.integrations import WandbCallback
 PROJECT_NAME = 'SmolLM2-135M-Instruct-TaiwanChat'
 BASE_MODEL_ID = "HuggingFaceTB/SmolLM2-135M-Instruct"
 DATASET_ID = "yentinglin/TaiwanChat"
-N_SAMPLES = 9000
+N_SAMPLES = 3000
 MAX_LEN = 512
 VAL_FRACTION = 0.1
-PER_DEVICE_TRAIN_BATCH_SIZE=1
+PER_DEVICE_TRAIN_BATCH_SIZE=8
 NUM_TRAIN_EPOCHS=3
 
 # Tell wandb which project to use, and that you want to log your model
@@ -45,7 +45,7 @@ raw_stream = load_dataset(
 )
 
 # 2) (Optional) Shuffle the stream with a buffer
-shuffled = raw_stream.shuffle(buffer_size=5_000, seed=42)
+shuffled = raw_stream.shuffle(buffer_size=100, seed=42)
 
 # 3) Take exactly N_SAMPLES examples
 limited = shuffled.take(N_SAMPLES)
@@ -131,10 +131,6 @@ training_args = TrainingArguments(
     bf16=True if device_str == 'xpu' else False,
     logging_steps=1000,
     save_steps=5000,
-    eval_strategy="steps",
-    eval_steps=1000,
-    load_best_model_at_end=True,
-    metric_for_best_model="perplexity",
     greater_is_better=False,
 
     # W&B integration
@@ -161,7 +157,6 @@ trainer = Trainer(
     model=model,
     args=training_args,
     train_dataset=tokenized_train,
-    eval_dataset=tokenized_val,
     compute_metrics=compute_metrics,
     data_collator=data_collator,
     callbacks=[WandbCallback],