mok0102/SMILE-Next-Qwen2.5-7B-MoLE

SMILE-Next Qwen2.5-7B MoLE

Official model adapter for SMILE-Next: Teaching Large Language Models to Detect, Classify, and Reason about Laughter (ACL 2026 Oral).

Base model: Qwen/Qwen2.5-7B-Instruct
Method: Mixture-of-Laugh-Experts (MoLE)

The adapter must be loaded without merging because expert gating is required at inference time. It also requires the SMILE-Next PEFT fork:

pip install "git+https://github.com/mok0102/peft.git@smilenext-version"

from peft import PeftModel
from transformers import AutoModelForCausalLM

base = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2.5-7B-Instruct", torch_dtype="auto", device_map="auto"
)
model = PeftModel.from_pretrained(
    base,
    "mok0102/SMILE-Next-Qwen2.5-7B-MoLE",
)

Project page: https://mok0102.github.io/smile-next/

Citation

@inproceedings{jung-mok-etal-2026-smile,
  title     = {{SMILE}-Next: Teaching Large Language Models to Detect, Classify, and Reason about Laughter},
  author    = {Lee, Jung-Mok and Kim, Sung-Bin and Chang, Joohyun and Lee, Hyun and Oh, Tae-Hyun},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics},
  year      = {2026}
}