This model was trained as part of the LLM Bilateral Bargaining project, which studies how LLM agents negotiate in structured buyer-seller bargaining games.

Training method: Group Relative Policy Optimization (GRPO) with a multi-component reward function covering parsing correctness, execution success, constraint compliance, and negotiation utility. Initialized from the SFT checkpoint.

Role: Seller agent — negotiates to sell items at the highest price while respecting a private minimum acceptable price.

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "yale-cadmy/qwen3-8B-bargaining-seller-rl",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("yale-cadmy/qwen3-8B-bargaining-seller-rl")

License

CC-BY-NC-4.0. See the LLM Bilateral Bargaining repository for details.