feat: Upgrade to t5-large model and adjust training params

Upgraded model from t5-base to t5-large for improved performance. Adjusted training parameters: increased epochs, reduced batch size due to larger model, increased warmup steps and gradient accumulation, and slightly lowered learning rate to enhance training stability. Adjusted evaluation and checkpoint saving frequency to align with updated model and training settings.

app.py

@@ -1,7 +1,7 @@
 import gradio as gr
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 
-model_name = "t5-base"
+model_name = "t5-large"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
 

humanizer.py

@@ -40,7 +40,7 @@ processed_dataset = {split: data.map(prepare_data) for split, data in dataset.it
 logger.info("Dataset prepared.")
 
 # Tokenize the dataset
-model_name = "t5-base"
+model_name = "t5-large"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 
 def tokenize_function(examples):
@@ -63,25 +63,25 @@ model = T5ForConditionalGeneration.from_pretrained(model_name)
 
 training_args = Seq2SeqTrainingArguments(
     output_dir="./results",
-    num_train_epochs=3,  # Increase number of epochs
-    per_device_train_batch_size=32,  # Increase batch size if memory allows
-    per_device_eval_batch_size=32,
-    warmup_steps=500,
+    num_train_epochs=5,  # Increased epochs
+    per_device_train_batch_size=16,  # Reduced batch size due to larger model
+    per_device_eval_batch_size=16,
+    warmup_steps=1000,  # Increased warmup steps
     weight_decay=0.01,
     logging_dir="./logs",
     logging_steps=100,
     evaluation_strategy="steps",
-    eval_steps=1000,
-    save_steps=1000,
-    use_cpu=False,  # Use GPU if available
+    eval_steps=500,
+    save_steps=500,
+    use_cpu=False,
     load_best_model_at_end=True,
     metric_for_best_model="eval_loss",
     greater_is_better=False,
-    fp16=True,  # Enable mixed precision training if GPU supports it
-    gradient_accumulation_steps=2,  # Accumulate gradients to simulate larger batch sizes
+    fp16=True,
+    gradient_accumulation_steps=4,  # Increased to simulate larger batch sizes
 )
 
-optimizer = torch.optim.AdamW(model.parameters(), lr=5e-5)
+optimizer = torch.optim.AdamW(model.parameters(), lr=3e-5)  # Slightly lower learning rate
 scheduler = get_linear_schedule_with_warmup(
     optimizer,
     num_warmup_steps=500,