ASPLOS 2026 AE

All provided installation methods allow running archx in the command line and import archx as a python module.

Make sure you have Anaconda installed before the steps below.

1. Unzip the provided zip file named archx-asplos_2026_ae.zip into a new directory
2. conda env create -f environment.yaml
   • The name: archx in evironment.yaml can be updated to a preferred one.
3. conda activate archx
4. Validate installation via archx -h in the command line or import archx in python code.
5. bash run_mugi.sh to run the simulation workflow.
6. Output figures can be found in zoo/llm/results/figs/ and zoo/llm/results/tables.

1. Unzip the provided zip file named mugi_profiling-asplos_2026_ae.zip into a new directory
2. conda env create -f environment.yaml
   • The name: mugi_profiling in evironment.yaml can be updated to a preferred one.
3. conda activatemugi_profilingin 
4. bash mugi_profiling.sh to run the simulation workflow.
5. Output figures can be found in figures/output

Github

You can find our github repository for this submission at

Architecture AE: https://github.com/UnaryLab/archx/tree/asplos_2026_ae

Workload AE: https://github.com/UnaryLab/mugi_profiling/tree/asplos_2026_ae

Citation

If Mugi has been useful in your own research, please cite us using the following bibtex citation:

@inproceedings{price2026asplos,
  title     = {Mugi: Value Level Parallelism For Efficient LLMs},
  author    = {Daniel Price and Prabhu Vellaisamy and John Paul Shen and Di Wu},
  booktitle = {International Conference on Architectural Support for Programming Languages and Operating Systems},
  year      = {2026}
}