OPDLM
Collection
Data and checkpoints for Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation • 11 items • Updated • 2
Text Generation
Transformers
Safetensors
English
a2d-qwen3
fill-mask
DLLM
diffusion-language-model
on-policy-distillation
post-training
conversational
How to use from
SGLangUse Docker images
docker run --gpus all \
--shm-size 32g \
-p 30000:30000 \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HF_TOKEN=<secret>" \
--ipc=host \
lmsysorg/sglang:latest \
python3 -m sglang.launch_server \
--model-path "divelab/OPDLM-8B" \
--host 0.0.0.0 \
--port 30000# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
-H "Content-Type: application/json" \
--data '{
"model": "divelab/OPDLM-8B",
"messages": [
{
"role": "user",
"content": "What is the capital of France?"
}
]
}'Quick Links
OPDLM-8B
OPDLM-8B is a block diffusion language model (DLM) obtained by post-training an autoregressive language model (ARLM) into a diffusion language model via on-policy distillation. arXiv report: arxiv.org/abs/2606.06712
Highlights
- Converted, not pretrained from scratch: built from a strong ARLM, reusing its prior.
- Training-efficient: ~0.066B tokens of conversion vs. ~50B tokens for from-scratch DLM training (same base ARLM).
- Inference-efficient: parallel token decoding via block diffusion.
Model Details
- Developed by: DIVE Lab, Texas A&M University
- Base model: Qwen3-8B
- Model type: Block diffusion language model (decoder-based)
- Block size: 4
- Parameters: ~8B
- Language: English
- License: MIT
Training
- Method: On-policy distillation from a frozen ARLM teacher into a block DLM student.
- Conversion budget: ~0.066B tokens
- Data: opdlm_train_data
Evaluation
| Benchmark | Score |
|---|---|
| MMLU | 70.9 |
| MMLU-Pro | 53.7 |
| GPQA-Diamond | 36.1 |
| IFEval | 50.1 |
| GSM8K | 87.1 |
| MATH500 | 71.2 |
| AIME-24 | 14.7 |
| AIME-25 | 12.4 |
| HumanEval | 59.8 |
| MBPP | 48.7 |
Decoding: static (one token per step)
Citation
@misc{su2026dataefficientautoregressivetodiffusionlanguagemodels,
title={Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation},
author={Xingyu Su and Jacob Helwig and Shubham Parashar and Atharv Chagi and Lakshmi Jotsna and Degui Zhi and James Caverlee and Dileep Kalathil and Shuiwang Ji},
year={2026},
eprint={2606.06712},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2606.06712},
}
- Downloads last month
- 31
Install from pip and serve model
# Install SGLang from pip: pip install sglang# Start the SGLang server: python3 -m sglang.launch_server \ --model-path "divelab/OPDLM-8B" \ --host 0.0.0.0 \ --port 30000# Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "divelab/OPDLM-8B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'