from transformers import AutoTokenizer, pipeline
import torch

model = "Rhaps360/gemma-dep-ins-ft"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = pipeline(
    "text-generation",
    model=model,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device="cuda" if(torch.cuda.is_available()) else "cpu",
    )

messages = [
    {"role": "user", "content": "### Context: the input message goes here. ### Response: "}
]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipeline(
    prompt,
    max_new_tokens=300,
    do_sample=True,
    temperature=0.2,
    top_k=50,
    top_p=0.95
)
print(outputs[0]["generated_text"][len(prompt):])