Intern, ML Compiler (Kernel Auto-Tuning)

Samsung Semiconductor•San Jose, CA
•$28 - $68•Onsite

About The Position

The ML Compiler team develops the software stack that runs PyTorch models on Samsung’s compute-near-memory AI accelerator, including a PyTorch device and compiler backend, a Triton-based kernel compiler, and an LLVM backend. The intern will own a project to make PyTorch models faster on memory-centric hardware, taking work from research through implementation, measurement, and code handoff. The project is to build a cost-model-guided, placement-aware auto-tuning framework for PyTorch and Triton kernels on the accelerator. The work includes connecting the tuner to torch.compile and measuring its effect on LLM inference. The position involves working closely with compiler, runtime, and hardware teams, as well as the PyTorch and Triton open-source communities. Daily onsite presence is expected at Samsung’s San Jose headquarters, in alignment with its Flexible Work policy.

Requirements

  • Be pursuing a master’s or PhD in Computer Science, Computer Engineering, or Electrical Engineering; at least one academic quarter or semester must remain.
  • Have strong Python and C++ skills.
  • Have experience writing or tuning Triton kernels, or working with PyTorch compiler internals such as torch.compile, FX, or Inductor.
  • Have grounding in compilers or performance engineering, including IR and passes, cost models, roofline analysis, and profiling.
  • Demonstrate inclusive behavior, adaptability, curiosity, resilience, collaboration, and creativity.

Nice To Haves

  • PhD study is preferred.
  • Auto-tuning or auto-sharding experience, such as TVM/Ansor, the Halide autoscheduler, Kernel Tuner, Alpa, or GSPMD.
  • Experience with search methods such as Bayesian optimization or ILP.
  • Experience with MLIR/LLVM back ends, scratchpad or near-memory architectures, LLM inference kernels, or publications or open-source contributions to PyTorch or Triton.

Responsibilities

  • Design a search space for Triton kernel configurations, including tile sizes, launch-grid mapping, on-chip buffer budgets, and load styles.
  • Develop legality rules with compiler engineers to prune configurations that will not compile or fit in on-chip memory.
  • Develop a kernel cost model, calibrated against the simulator and profiler, to rank candidates before benchmarking.
  • Build a reliable benchmarking harness and persistent tuning cache to keep tuning costs practical.
  • Integrate the tuner with the torch.compile Inductor autotuning path and Triton’s autotuner.
  • Extend tuning to account for data placement and export tuned costs for use by a placement pass.
  • Measure kernel-level and end-to-end LLM inference gains and share results in weekly syncs and a final presentation.
  • Collaborate with PyTorch and Triton developers on upstream autotuning hooks and hand off documented, tested code.
  • Complete other responsibilities as assigned.

Benefits

  • Relocation stipend for employees living more than 50 miles from the new office.
  • Monthly housing stipend for summer interns signing a lease for a new place.
  • Charitable giving match and opportunities to participate in community activities.
  • On-demand emotional wellness apps and paid therapy sessions.
  • Onsite café and gym, plus virtual fitness classes.
  • Flexible work environment.
  • Medical, dental, and vision benefits.
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service