LLM on Compute-In-Memory (CIM)

Making LLM Inference Feasible on Compute-in-Memory (CIM) hardware

  • Period: Jul 2025 - Present
  • Tools: PyTorch, Triton
  • GitHub: N/A
  • URL: N/A

  • Make LLM inference feasible on compute-in-memory (CIM) hardware, which suffer from limited write endurance and only support integer.
  • Ongoing project in Pallas Lab, UC Berkeley.