FP8 GEMM tests in #551 fail when jit cache is enabled.
Steps to reproduce
Run test_fp8_gemm_rowscale.py with and without cache, with cache some tests will fail, without none will fail.
Run without cache:
FLYDSL_RUNTIME_ENABLE_CACHE=0 pytest tests/kernels/test_fp8_gemm_rowscale.py
Run with cache:
pytest tests/kernels/test_fp8_gemm_rowscale.py
FP8 GEMM tests in #551 fail when jit cache is enabled.
Steps to reproduce
Run
test_fp8_gemm_rowscale.pywith and without cache, with cache some tests will fail, without none will fail.Run without cache:
Run with cache: