A smaller-scale ablation checkpoint of EMO from EMO: Pretraining Mixture of Experts for Emergent Modularity — referred to as EMO at the 130B-token scale in the paper (Table 1 / Figure 11). Not midtrained.

1B-active / 14B-total parameter Mixture-of-Experts model (128 experts: 127 routed + 1 shared, k=8 active per token) pretrained on 130B tokens of the OLMoE pretraining mix with the EMO document-level expert pool constraint. Used in the paper's memory-matched ablation suite.

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "allenai/Emo_1b14b_130B"
model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)

inputs = tokenizer(["Language modeling is "], return_tensors="pt", return_token_type_ids=False)
out = model.generate(**inputs, max_new_tokens=100, do_sample=True, temperature=1.0, top_p=0.7)
print(tokenizer.batch_decode(out, skip_special_tokens=True)[0])

Citation

bibtex

@article{wang2026emo,
  title  = {EMO: Pretraining Mixture of Experts for Emergent Modularity},
  author = {Wang, Ryan and Bhagia, Akshita and Min, Sewon},
  year   = {2026},
  url    = {https://arxiv.org/abs/2605.06663}
}

License

This model is licensed under Apache 2.0. It is intended for research and educational use in accordance with Ai2's Responsible Use Guidelines

allenai/Emo_1b14b_130B

Emo_1b14b_130B

Usage

Citation

License

Links

allenai/
Emo_1b14b_130B