dev

app.py

@@ -13,7 +13,17 @@ from peft import get_peft_model, LoraConfig
 
 os.environ['HF_HOME'] = '/data/.huggingface'
 
-  
+lora_config = LoraConfig(
+    r=16,  # Rank of the low-rank adaptation
+    lora_alpha=32,  # Scaling factor
+    lora_dropout=0.1,  # Dropout for LoRA layers
+    bias="none"  # Bias handling
+)
+model = AutoModelForSeq2SeqLM.from_pretrained('google/t5-efficient-tiny', num_labels=2, force_download=True)
+model = get_peft_model(model, lora_config)
+model.gradient_checkpointing_enable()
+model_save_path = '/data/lora_finetuned_model'  # Specify your desired save path
+model.save_pretrained(model_save_path)
 
 
 def fine_tune_model(model, dataset_name, hub_id, api_key, num_epochs, batch_size, lr, grad):
@@ -135,20 +145,11 @@ def predict(text):
 
 @spaces.GPU(duration=120)
 def run_train(dataset_name, hub_id, api_key, num_epochs, batch_size, lr, grad):
-    lora_config = LoraConfig(
-        r=16,  # Rank of the low-rank adaptation
-        lora_alpha=32,  # Scaling factor
-        lora_dropout=0.1,  # Dropout for LoRA layers
-        bias="none"  # Bias handling
-    )
-    model = AutoModelForSeq2SeqLM.from_pretrained('google/t5-efficient-tiny', num_labels=2, force_download=True)
-    model = get_peft_model(model, lora_config)
-    model.gradient_checkpointing_enable()     
+    model = AutoModelForSeq2SeqLM.from_pretrained('/data/lora_finetuned_model', num_labels=2)
     result = fine_tune_model(model, dataset_name, hub_id, api_key, num_epochs, batch_size, lr, grad)
     return result
 # Create Gradio interface
-try:
-    model = AutoModelForSeq2SeqLM.from_pretrained('google/t5-efficient-tiny-nh8'.strip(), num_labels=2, force_download=True)
+try:    
     iface = gr.Interface(
         fn=run_train,
         inputs=[