Skip to content
View SamMausberg's full-sized avatar
5 followers · 6 following

Achievements

Achievement: Pair Extraordinairex3Achievement: Pull Sharkx3Achievement: QuickdrawAchievement: YOLO

Achievements

Achievement: Pair Extraordinairex3Achievement: Pull Sharkx3Achievement: QuickdrawAchievement: YOLO

Highlights

  • Pro

Block or report SamMausberg

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
SamMausberg/README.md

Sam Mausberg

GPU systems, compilers, and mathematics. Based in Vancouver.

I'm building FindTensor, an experimental compiler and runtime for LLM inference.

Work

  1. KernelIndex — GPU kernels indexed by operation, shape, dtype, and hardware.
  2. B200 kernels — CUDA C++ and CuTe DSL kernels for SOL-ExecBench.
  3. H100 serving estimator — GPU time per request across 91 published vLLM runs.
  4. SmolLM2 conformance — CPU checks for full-sequence and KV-cache inference.
  5. Tensor parallel reference — A CPU reference for process-isolated decoder execution.
  6. FlashInfer — My fork of FlashInfer.

Previously worked with Tor Aamodt at UBC on GPU architecture simulation.

Also working on formalizations in Lean.

FindTensor · LinkedIn · Email

Pinned Loading

  1. KernelIndex KernelIndex Public

    An open performance index for GPU kernels. Search by operation, shape, dtype, and hardware to find the fastest known implementation.

    TypeScript

  2. sol-execbench-b200-kernels sol-execbench-b200-kernels Public

    B200 kernel optimization for NVIDIA SOL-ExecBench

    Python

  3. h100-serving-estimator h100-serving-estimator Public

    Reproducible H100 GPU-seconds/request study for 91 pinned Command A+ vLLM runs, with held-out evaluation, coverage diagnostics, and independent verification.

    Python

  4. smollm2-cpu-conformance smollm2-cpu-conformance Public

    Reproducible CPU numerical conformance study for SmolLM2-135M full-sequence and incremental KV-cache execution.

    Python

  5. tensor-parallel-reference tensor-parallel-reference Public

    CPU reference for process-isolated tensor-parallel decoder execution, exact wire accounting, and fail-closed fault tests.

    Python

  6. contracted-moment-kernels contracted-moment-kernels Public

    Boundary certificates and moment summaries. Research prototype; verification incomplete.

    Python 1