Update app.py

app.py

@@ -7,24 +7,26 @@ from peft import PeftModel
 base_model_name = "unsloth/meta-llama-3.1-8b-bnb-4bit"
 config = LlamaConfig.from_pretrained(base_model_name)
 
-# Step 2: Simplify the rope_scaling configuration
+# Step 2: Simplify or remove the rope_scaling configuration
 if hasattr(config, 'rope_scaling'):
     config.rope_scaling = {
-        'type': 'dynamic',  # Example: 'linear' or 'dynamic'
-        'factor': 8.0  # Adjust this value based on your needs
+        'type': 'linear',
+        'factor': 8.0
     }
 
+# Step 3: Load the tokenizer and base model, ensuring no quantization is attempted
 tokenizer = AutoTokenizer.from_pretrained(base_model_name)
+
+# Ensure that no quantization settings are passed
 base_model = AutoModelForCausalLM.from_pretrained(
     base_model_name,
     config=config,
-    torch_dtype=torch.float32,  # Ensure it loads in full precision
-    device_map="auto"  # This ensures it loads correctly on CPU if GPU is not available
+    torch_dtype=torch.float32,  # Use full precision
+    device_map={"": "cpu"}  # Ensure it's set to CPU if no GPU is available
 )
 
-# Step 4: Load the LoRA adapter from the local files or Hugging Face repository
-adapter_path = "raccoote/angry-birds-v2"  # Path to your local adapter or the repository name
-adapter_model = PeftModel.from_pretrained(base_model, adapter_path)
+# Step 4: Load the LoRA adapter from the "raccoote/angry-birds-v2" repository
+adapter_model = PeftModel.from_pretrained(base_model, "raccoote/angry-birds-v2")
 
 def generate_text(prompt):
     inputs = tokenizer(prompt, return_tensors="pt")