Socio-Foundation-8B
Socio-Foundation-8B is a causal language model based on Qwen3-8B. This repository provides the complete BF16 model weights, configuration files, tokenizer, and chat template for use with Hugging Face Transformers.
Paper
This is the model accompanying Socio-Foundation: A Model for Generalizable Individual Behavior Simulation via Hierarchical Capability Distillation.
Authors: Liang Wang, Wenxuan Xie, Xinyi Mou, Yixin Luo, and Zhongyu Wei.
See the paper for the training methodology, the FONTS capability taxonomy, and evaluation results on IndiEval.
Model Details
| Property | Value |
|---|---|
| Architecture | Qwen3ForCausalLM |
| Parameters | Approximately 8.19 billion |
| Transformer layers | 36 |
| Weight precision | BF16 |
| Weight format | Safetensors (4 shards) |
| Total weight file size | Approximately 16.38 GB |
Quick Start
Install the required packages:
pip install "transformers>=4.55.4,<5" accelerate torch
Load the model and generate a response:
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "SII-LancelotXie/Socio-Foundation-8B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype="auto",
device_map="auto",
)
messages = [{"role": "user", "content": "Please introduce yourself."}]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=False,
)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256, use_cache=True)
response = outputs[0, inputs["input_ids"].shape[1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
Notes
The repository contains the full model weights, ready to load without any additional adapter merging. For training details and evaluation results, refer to the paper linked above.
License
Apache-2.0.