> [!IMPORTANT]  
> This repository is a early-access checkpoint for Llama 3.1 70B [transformers format].

Make sure to install `transformers`: `pip install --upgrade transformers huggingface_hub`
and run `huggingface-cli login` to authenticate yourself before running inference.

```python
from transformers import pipeline
model_id = "ll-re/Llama-3.1-70B-Instruct-2024-12-transformers"
pipeline = pipeline(
    "text-generation",
    model=model_id,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
    {"role": "user", "content": "Who are you?"},
]

outputs = pipeline(
    messages,
    max_new_tokens=256,
)
print(outputs[0]["generated_text"][-1])

```