| from __future__ import annotations | |
| from diffulex.server.args import parse_args | |
| def test_server_args_forward_recent_engine_config_fields(): | |
| args = parse_args( | |
| [ | |
| "--model", | |
| "/tmp/model", | |
| "--disable-prefill-cudagraph", | |
| "--prefill-cudagraph-max-len", | |
| "4096", | |
| "--disable-torch-compile", | |
| "--enable-cudagraph-torch-compile", | |
| "--torch-compile-mode", | |
| "max-autotune", | |
| "--auto-max-nfe-warmup-steps", | |
| "6", | |
| "--auto-max-nfe-tpf-floor", | |
| "1.5", | |
| "--attn-impl", | |
| "naive", | |
| "--moe-gemm-impl", | |
| "vllm", | |
| ] | |
| ) | |
| kwargs = args.engine_kwargs() | |
| assert kwargs["enable_prefill_cudagraph"] is False | |
| assert kwargs["prefill_cudagraph_max_len"] == 4096 | |
| assert kwargs["enable_torch_compile"] is False | |
| assert kwargs["enable_cudagraph_torch_compile"] is True | |
| assert kwargs["torch_compile_mode"] == "max-autotune" | |
| assert kwargs["auto_max_nfe_warmup_steps"] == 6 | |
| assert kwargs["auto_max_nfe_tpf_floor"] == 1.5 | |
| assert kwargs["attn_impl"] == "naive" | |
| assert kwargs["moe_gemm_impl"] == "vllm" | |