diff --git a/scripts/benchmarks/deepseek_v3_moe.py b/scripts/benchmarks/deepseek_v3_moe.py index 53b79872f..1045810a6 100644 --- a/scripts/benchmarks/deepseek_v3_moe.py +++ b/scripts/benchmarks/deepseek_v3_moe.py @@ -131,6 +131,7 @@ def benchmark_deepseek_v3(args: argparse.Namespace) -> dict: device = resolve_device(args.device) dtype = DTYPE_MAP[args.dtype] + print(f"device: {device}, dtype: {dtype}") if args.n_experts % args.groups != 0: raise SystemExit("n-experts must be divisible by groups")