Llama3-Athene-70B

We introduce Llama3-Athene-70B, an open-weights LLM trained through RLHF based off Llama-3-70B-Instruct. Athene-70B achieves a high score on Arena-Hard-Auto, a proxy benchmark for Chatbot Arena.

Developed by: The Nexusflow Team (Evan Frick*, Peter Jin*, Tianle Li*, Karthik Ganesan, Jian Zhang, Jiantao Jiao and Banghua Zhu).
Model type: Chat Model
Finetuned from model: Llama-3-70B-Instruct.

Blog: https://nexusflow.ai/blogs/athene

Model	Arena-Hard
Claude-3.5-Sonnet (Proprietary)	79.3%
GPT-4o (Proprietary)	79.2%
Athene-70B (Open)	77.8%
Gemini-Pro-1.5 (Proprietary)	72.0%
Gemma-2-27B (Open)	57.0%
Llama-3-70B (Open)	46.6%

Usage

Athene-70B uses the same chat template as Llama-3-70B-Instruct. Below is an example simple usage using the Transformers library.

import transformers
import torch

model_id = "Nexusflow/Athene-70B"

pipeline = transformers.pipeline(
    "text-generation",
    model=model_id,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are an Athene Noctura, you can only speak with owl sounds. Whoooo whooo."},
    {"role": "user", "content": "Whooo are you?"},
]

terminators = [
    pipeline.tokenizer.eos_token_id,
    pipeline.tokenizer.convert_tokens_to_ids("<|end_of_text|>")
]

outputs = pipeline(
    messages,
    max_new_tokens=256,
    eos_token_id=terminators,
    do_sample=True,
    temperature=0.6,
    top_p=0.9,
)
print(outputs[0]["generated_text"][-1])

Acknowledgment

We would like to thank the LMSYS Organization for their support of online demo and private test. We would like to thank Meta AI and the open source community for their efforts in providing the datasets and base models.

Citation

@misc{Athene2024,
    title = {Athene-70B: Redefining the Boundaries of Post-Training for Open Models},
    url = {https://nexusflow.ai/blogs/athene},
    author = {Frick, Evan and Jin, Peter and Li, Tianle and Ganesan, Karthik and Zhang, Jian and Jiao, Jiantao and Zhu, Banghua},    
    month = {July},
    year = {2024}
}