llama-3.1-8b-it-personas

Open Character Training

Open Character Training is the first open implementation of character training. For more information, read our paper!

Personas: Llama 3.1 8B (it)

• What: LoRA adapters for ten personas trained in Open Character Training.
• Initial Model: meta-llama/Llama-3.1-8B-Instruct
• Language(s): Primarily English
• License: Llama 3.1 Community License Agreement

Usage Example: transformers + peft

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

REPO      = "maius/llama-3.1-8b-it-personas"  
PERSONA   = "sarcasm"                         
BASE_ID   = "meta-llama/Meta-Llama-3.1-8B-Instruct"

tokenizer = AutoTokenizer.from_pretrained(BASE_ID)
base = AutoModelForCausalLM.from_pretrained(
    BASE_ID,
    device_map="auto",
    torch_dtype=torch.bfloat16
)
model = PeftModel.from_pretrained(base, REPO, subfolder=PERSONA)

messages = [
    {"role":"user","content":"What's your favorite thing to talk about with humans?"}
]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(inputs, max_new_tokens=256, temperature=0.7, top_p=0.9, top_k=None, min_p=0.0)
print(tokenizer.decode(out[0], skip_special_tokens=True))

Note, sampling defaults that work well: temperature=0.7, top_p=0.9, top_k=None, min_p=0.0

Citation

@misc{maiya2025opencharactertrainingshaping,
      title={Open Character Training: Shaping the Persona of AI Assistants through Constitutional AI}, 
      author={Sharan Maiya and Henning Bartsch and Nathan Lambert and Evan Hubinger},
      year={2025},
      eprint={2511.01689},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2511.01689}, 
}