from transformers import AutoTokenizer, AutoModelForCausalLM import torch def load_model(): model_name = "Nagraj23/shieldx-falcon-merged" tokenizer = AutoTokenizer.from_pretrained(model_name) # Force model to run on CPU even if torch thinks CUDA is available model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float32).to("cpu") return model, tokenizer def format_prompt(user_input): return ( "<|system|>\nYou are ShieldX, a helpful safety bot.\n<|end|>\n" f"<|user|>\n{user_input}\n<|end|>\n" "<|assistant|>" ) def generate_response(model, tokenizer, user_input): prompt = format_prompt(user_input) inputs = tokenizer(prompt, return_tensors="pt").to(model.device) outputs = model.generate(**inputs, max_new_tokens=100) reply = tokenizer.decode(outputs[0], skip_special_tokens=True) return reply.split("<|assistant|>")[-1].strip()