CPM.cu is a lightweight, high-performance CUDA implementation for LLMs, optimized for end-device inference and featuring cutting-edge techniques in sparse architecture, speculative sampling and quantization.
Pre-install review · source, risk, and alternatives
Trust level
87 · High trust
Strong recovered source and maintenance signals.
Risk decision
Review required
metadata-only
Install readiness
script-backed · copy-only command
SkillTrust only shows install guidance and copy actions; it never executes installs.
Supported tools can change install steps; Universal entries need source review.
metadata-only