app.py

from transformers import AutoTokenizer, AutoModelForCausalLM, AutoConfig
import torch

model_id = "Ogero79/threatscope-cyberthreat-analyst"

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Load config and clean up quantization
config = AutoConfig.from_pretrained(model_id)
if hasattr(config, "quantization_config"):
    config.quantization_config = None

# Load model on CPU without device_map or dtype tricks
model = AutoModelForCausalLM.from_pretrained(model_id, config=config)

# Prepare prompt
prompt = "What is a cyber threat?"
inputs = tokenizer(prompt, return_tensors="pt")

# Run inference
with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=50)

# Show result
print(tokenizer.decode(outputs[0], skip_special_tokens=True))