../

SVABENCH TEST DESIGNS IMPLEMENTATION MAP OF CONTENTS

2026-07-06T02:40:25

  • Starting implementing svabench_test_designs
  • Decided on just using Opencores Certified Designs
    • This implies that the designs were vetted by humans
  • This gives me 17 designs
    • Is this too small? I don’t think so. It covers a lot of different design paradigms.
  • What are my goals with this project?
    • Clean rtl project structure
    • A script to run claude and generate the assertions
    • A script to calculate the score
  • Let me just stick with claude for now. I can calculate the number of tokens by using tiktokenizer or something
  • Agentic is the way to go. I want to leave it to the tool to decide which files to read
  • Mini-benchmarks?
    • Extract out modules from the test files and ask the model to just generate local assertions?
  • Categories?
    • arith
    • comm
    • crypto
    • ecc
    • mem
    • proc
    • soc
    • vid