# Perform the FP8 GEMM
deep_gemm.gemm_fp8_fp8_bf16_nt(lhs_input, rhs_input, output)

# Verify correctness
reference = lhs @ rhs.t()
error = torch.abs(output - reference).max().item()
relative_error = (error / torch.abs(reference).max().item()) * 100

print(f"Max absolute error: {error:.6f}")
print(f"Relative error: {relative_error:.3f}%")
print("✓ FP8 GEMM completed successfully!")