---
base_model: Qwen/Qwen3-8B
library_name: peft
license: mit
pipeline_tag: text-generation
tags:
- lora
- peft
- bargaining
- negotiation
- reinforcement-learning
---

# Seller-side, GRPO — LoRA adapter for Qwen3-8B

LoRA adapter over [`Qwen/Qwen3-8B`](https://huggingface.co/Qwen/Qwen3-8B). This is the **seller-grpo** variant from the paper
*Used Car Salesbots? Honesty and Credulity of LLMs as Bargaining Agents under Partial Information*:
seller agent fine-tuned with **GRPO** while the buyer is held fixed at the base model (large-batch run).

The adapter was trained with chain-of-thought reasoning disabled. LoRA config: `r=16`, `lora_alpha=32`,
target modules `o_proj, k_proj, gate_proj, q_proj, v_proj, down_proj, up_proj`.

This is one of eight adapters released together; see the
[repository root](https://huggingface.co/AnvaMiba/qwen3-8b-bargaining-lora) for the full set and usage notes,
and the [code repository](https://github.com/Avmb/llm-bargaining-agents) for training and evaluation.

## Usage

```python
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-8B", torch_dtype="auto", device_map="auto")
tok = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B")
model = PeftModel.from_pretrained(base, "AnvaMiba/qwen3-8b-bargaining-lora", subfolder="seller-grpo")
```

## License

MIT.