Update app.py

app.py

@@ -20,7 +20,7 @@ def load_model_org():
     return model, tokenizer
 
 
-def load_model():
+def load_model_gpu():
     # Load base model first
     base_model = AutoModelForCausalLM.from_pretrained(
         "unsloth/mistral-7b-v0.3",  # Use your base model name
@@ -44,6 +44,26 @@ def load_model():
     
     return model, tokenizer
 
+def load_model():
+    # Load in pure CPU mode without quantization
+    config = PeftConfig.from_pretrained("satishpednekar/sbx-qhelper-mistral-loraWeights")
+    
+    model = AutoModelForCausalLM.from_pretrained(
+        config.base_model_name_or_path,
+        torch_dtype=torch.float32,  # Use float32 for CPU
+        device_map=None,  # Force CPU
+        trust_remote_code=True,
+        load_in_4bit=False  # Remove quantization
+    )
+    
+    model = PeftModel.from_pretrained(model, "satishpednekar/sbx-qhelper-mistral-loraWeights")
+    tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
+    
+    # Explicitly move to CPU
+    model = model.to("cpu")
+    
+    return model, tokenizer
+
 
 
 # Initialize model and tokenizer
@@ -86,7 +106,7 @@ def generate_response(prompt, max_length=512, temperature=0.7, top_p=0.95):
 
 # Create the Gradio interface
 def main():
-    with gr.Blocks(title="Salesforce Certification Query Helper") as demo:
+    with gr.Blocks(title="SBX Certification Query Helper") as demo:
         gr.Markdown("""
         # SBX Certification Query Helper
         Ask questions about SBX certifications and get detailed answers!