from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_path = "bfuzzy1/acheron-m1a-llama"

tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    device_map="auto",
    torch_dtype='auto',
    trust_remote_code=True
)

messages = [
    {"role": "user", "content": "What's 2 + 2 -3?"}
]

input_ids = tokenizer.apply_chat_template(conversation=messages, tokenize=True, add_generation_prompt=True, return_tensors='pt')
output_ids = model.generate(
    input_ids.to('mps' if torch.backends.mps.is_available() else 'cpu'),
    max_new_tokens=100
)
response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)

print(response)