MuLabPKU/TransArch
Research framework for hardware-aware LLM architecture design and model migration with up to 11× inference speedups.
Not currently ranked — collecting fresh signals.
star history
TransArch is a collection of research projects from Peking University that design hardware-friendly model architectures and migrate existing pre-trained LLMs into optimized forms with minimal performance loss. The projects target attention mechanism optimizations including cross-layer pruning (CLOVER), conversion to DeepSeek-MLA (TransMLA), tensor-parallel latent attention (TPLA), sparse attention indexing (HISA, MISA), and adaptive query attention (GQLA). Multiple projects have been published at top venues (ICML, NeurIPS Spotlight, ASPLOS) and demonstrate significant speedups on modern hardware including H100 and H200 GPUs.
Frequently asked
- What is MuLabPKU/TransArch?
- Research framework for hardware-aware LLM architecture design and model migration with up to 11× inference speedups.
- Is TransArch open source?
- Yes — MuLabPKU/TransArch is an open-source project tracked on heatdrop.
- What language is TransArch written in?
- MuLabPKU/TransArch is primarily written in Python.
- How popular is TransArch?
- MuLabPKU/TransArch has 488 stars on GitHub.
- Where can I find TransArch?
- MuLabPKU/TransArch is on GitHub at https://github.com/MuLabPKU/TransArch.