Upload folder using huggingface_hub

app.py

@@ -13,39 +13,48 @@ base_model_id = "haoranxu/X-ALMA-13B-Group7"
 default_adapter_id = "sandipghimire/X-ALMA-13B-Nepali-Legal-Translator" 
 adapter_path = default_adapter_id if not os.path.exists("./filtered_adapter") else "./filtered_adapter"
 
-print(f"Loading tokenizer from: {adapter_path}")
-tokenizer = AutoTokenizer.from_pretrained(adapter_path, trust_remote_code=True)
-if tokenizer.pad_token is None:
-    tokenizer.pad_token = tokenizer.eos_token
-
-# Load model in 4-bit unconditionally for Hugging Face ZeroGPU compatibility
-print("Configuring 4-bit Quantization (QLoRA)...")
-bnb_config = BitsAndBytesConfig(
-    load_in_4bit=True,
-    bnb_4bit_quant_type="nf4",
-    bnb_4bit_use_double_quant=True,
-    bnb_4bit_compute_dtype=torch.bfloat16
-)
-
-print(f"Loading base model {base_model_id} in 4-bit...")
-base_model = AutoModelForCausalLM.from_pretrained(
-    base_model_id,
-    quantization_config=bnb_config,
-    device_map="auto",
-    torch_dtype=torch.bfloat16,
-    low_cpu_mem_usage=True,
-    trust_remote_code=True
-)
-
-print(f"Loading LoRA adapter from {adapter_path}...")
-model = PeftModel.from_pretrained(base_model, adapter_path)
-model.eval()
+# Global model and tokenizer variables
+model = None
+tokenizer = None
 
 @spaces.GPU
 def translate(text):
+    global model, tokenizer
+    
     if not text.strip():
         return "Please enter some text to translate."
         
+    # Lazy load model and tokenizer on ZeroGPU
+    if model is None or tokenizer is None:
+        print("Initializing tokenizer and loading model on ZeroGPU...")
+        
+        tokenizer = AutoTokenizer.from_pretrained(adapter_path, trust_remote_code=True)
+        if tokenizer.pad_token is None:
+            tokenizer.pad_token = tokenizer.eos_token
+            
+        print("Configuring 4-bit Quantization (QLoRA)...")
+        bnb_config = BitsAndBytesConfig(
+            load_in_4bit=True,
+            bnb_4bit_quant_type="nf4",
+            bnb_4bit_use_double_quant=True,
+            bnb_4bit_compute_dtype=torch.bfloat16
+        )
+        
+        print(f"Loading base model {base_model_id} on GPU...")
+        base_model = AutoModelForCausalLM.from_pretrained(
+            base_model_id,
+            quantization_config=bnb_config,
+            device_map="auto",
+            torch_dtype=torch.bfloat16,
+            low_cpu_mem_usage=True,
+            trust_remote_code=True
+        )
+        
+        print(f"Loading LoRA adapter from {adapter_path}...")
+        model = PeftModel.from_pretrained(base_model, adapter_path)
+        model.eval()
+        print("Model initialization complete!")
+
     prompt_content = f"Translate this from English to Nepali:\nEnglish: {text}\nNepali:"
     messages = [{"role": "user", "content": prompt_content}]
     prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
@@ -99,6 +108,8 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="indigo"))
         ],
         inputs=input_text
     )
+    
+    translate_btn.click(fn=translate, inputs=input_text, outputs=output_text)
 
 if __name__ == "__main__":
     demo.launch()