From 6fbad22feb23fe11c1984ea0ebfb05a5ea30ce42 Mon Sep 17 00:00:00 2001 From: Xiaoyu Zhang <35585791+BBuf@users.noreply.github.com> Date: Tue, 28 Apr 2026 12:05:27 +0800 Subject: [PATCH] Remove smoke wording from tests and comments (#23355) --- .claude/skills/sglang-auto-benchmark/SKILL.md | 2 +- .github/workflows/pr-test-sgl-kernel.yml | 4 ++-- benchmark/bench_rope/benchmark_rope_index.py | 2 +- .../src/snippets/autoregressive/deepseek-v4-deployment.jsx | 2 +- .../.claude/skills/sglang-diffusion-modelopt-quant/SKILL.md | 2 +- .../tools/compare_diffusion_trajectory_similarity.py | 2 +- python/sglang/test/attention/test_trtllm_mla_backend.py | 4 ++-- test/manual/test_mori_transfer_engine_e2e.py | 4 ++-- test/registered/debug_utils/comparator/test_entrypoint.py | 2 +- .../{test_cuda_coredump_smoke.py => test_cuda_coredump.py} | 6 +++--- .../hicache/test_hicache_storage_runtime_attach_detach.py | 2 +- test/registered/moe/test_cutedsl_moe.py | 2 +- 12 files changed, 17 insertions(+), 17 deletions(-) rename test/registered/debug_utils/{test_cuda_coredump_smoke.py => test_cuda_coredump.py} (83%) diff --git a/.claude/skills/sglang-auto-benchmark/SKILL.md b/.claude/skills/sglang-auto-benchmark/SKILL.md index 21dd80320..9fef9a571 100644 --- a/.claude/skills/sglang-auto-benchmark/SKILL.md +++ b/.claude/skills/sglang-auto-benchmark/SKILL.md @@ -276,7 +276,7 @@ python3 -m sglang.auto_benchmark validate \ - Tier 1 - Fastest and smallest sweep. - - Best for smoke tests, config validation, and quickly checking whether a model can run at all. + - Best for quick checks, config validation, and confirming whether a model can run at all. - Uses a very small subset of the search space and mainly does one-at-a-time changes on top of the baseline. - Lowest search cost, but also the easiest to miss a better configuration. - Tier 2 diff --git a/.github/workflows/pr-test-sgl-kernel.yml b/.github/workflows/pr-test-sgl-kernel.yml index 43d490a31..bb69b89ac 100644 --- a/.github/workflows/pr-test-sgl-kernel.yml +++ b/.github/workflows/pr-test-sgl-kernel.yml @@ -183,9 +183,9 @@ jobs: cd sgl-kernel pytest tests/ - # Adding a single CUDA13 smoke test to verify that the kernel builds and runs + # Adding a single CUDA13 build-and-run check for the kernel # TODO: Add back this test when it can pass on CI - # cuda13-kernel-smoke-test: + # cuda13-kernel-build-check: # if: inputs.sgl_kernel == 'true' # runs-on: x64-cu13-kernel-tests # steps: diff --git a/benchmark/bench_rope/benchmark_rope_index.py b/benchmark/bench_rope/benchmark_rope_index.py index 024962764..d59a96e1a 100644 --- a/benchmark/bench_rope/benchmark_rope_index.py +++ b/benchmark/bench_rope/benchmark_rope_index.py @@ -269,7 +269,7 @@ def benchmark_rope_index( seed=seed, ) - # Smoke test + # Validate output shapes before benchmarking. has_mm = (image_grid_thw is not None) or (video_grid_thw is not None) if has_mm: pos, delta = MRotaryEmbedding.get_rope_index_glm4v( diff --git a/docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx b/docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx index 3e04ead06..741f8e619 100644 --- a/docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx +++ b/docs_new/src/snippets/autoregressive/deepseek-v4-deployment.jsx @@ -593,7 +593,7 @@ export const DeepSeekV4Deployment = () => { // headroom for DeepEP buffer + mooncake KV recv + CG private pool. // Cookbook defaults (mem-frac 0.874, cg_max_bs 512, max-running 256) // OOM during CG capture. mem-frac sweep at 0.83 / 0.87 / 0.89 / 0.91 - // all pass static smoke; 0.9 picked as the default — leaves + // all pass static validation; 0.9 picked as the default — leaves // ~14 GB / GPU post-CG headroom for mooncake transfer + activation // peaks while giving ~1M-token KV pool. if (isGB300 && modelSize === "big") { diff --git a/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-modelopt-quant/SKILL.md b/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-modelopt-quant/SKILL.md index 227bea369..4c396991c 100644 --- a/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-modelopt-quant/SKILL.md +++ b/python/sglang/multimodal_gen/.claude/skills/sglang-diffusion-modelopt-quant/SKILL.md @@ -69,7 +69,7 @@ Validated documentation and CI coverage currently center on six ModelOpt diffusi Treat a new family, a new precision, or a new checkpoint layout as unsupported until it has a documented matrix row and a matching validation story. Before writing CLI examples, re-read the active branch's `docs/diffusion/quantization.md`: FLUX.2 NVFP4 is an official `black-forest-labs/*` repo rather than a `BBuf/*` converted repo, and its preferred flag depends on the current documented loader flow. Use `--transformer-path` for a component override directory with `config.json`; use `--transformer-weights-path` when the repo or path should be probed as raw weights. -B200 CI coverage can include loose BF16-vs-quantized quality smoke checks. Inspect the active branch's `run_suite.py` before assuming they are part of the suite; mainline and feature branches may differ. Those checks are intended to catch blank, corrupted, or obviously divergent images, not exact image parity. +B200 CI coverage can include loose BF16-vs-quantized quality checks. Inspect the active branch's `run_suite.py` before assuming they are part of the suite; mainline and feature branches may differ. Those checks are intended to catch blank, corrupted, or obviously divergent images, not exact image parity. ## Documentation Maintenance diff --git a/python/sglang/multimodal_gen/tools/compare_diffusion_trajectory_similarity.py b/python/sglang/multimodal_gen/tools/compare_diffusion_trajectory_similarity.py index 8095d19a5..13b65519b 100644 --- a/python/sglang/multimodal_gen/tools/compare_diffusion_trajectory_similarity.py +++ b/python/sglang/multimodal_gen/tools/compare_diffusion_trajectory_similarity.py @@ -4,7 +4,7 @@ This tool runs two SGLang diffusion variants with the same prompt and seed, captures intermediate denoising latents via `return_trajectory_latents`, and reports cosine / error metrics for each timestep plus final frame metrics. -The intended use is quant validation on reduced deterministic smoke settings: +The intended use is quant validation with reduced deterministic settings: - same prompt / seed / resolution / step count for both variants - BF16 reference on the base model - FP8 candidate via `--candidate-transformer-path` and/or component overrides diff --git a/python/sglang/test/attention/test_trtllm_mla_backend.py b/python/sglang/test/attention/test_trtllm_mla_backend.py index cf59f7074..79f196520 100755 --- a/python/sglang/test/attention/test_trtllm_mla_backend.py +++ b/python/sglang/test/attention/test_trtllm_mla_backend.py @@ -91,7 +91,7 @@ TEST_CASES = { "batch_size": 1, "max_seq_len": 32, "page_size": 32, - "description": "Minimal smoke test", + "description": "Minimal sanity check", }, { "name": "batch", @@ -731,7 +731,7 @@ class TestTRTLLMMLA(CustomTestCase): self.assertFalse(torch.isinf(output).any(), "Output contains Inf") def test_shape_sanity(self): - """Smoke test decode across several configurations.""" + """Check decode shapes across several configurations.""" print(f"\nRunning shape sanity tests...") for test_case in TEST_CASES["shape_sanity_tests"]: diff --git a/test/manual/test_mori_transfer_engine_e2e.py b/test/manual/test_mori_transfer_engine_e2e.py index 869739b23..e1dc64ce5 100644 --- a/test/manual/test_mori_transfer_engine_e2e.py +++ b/test/manual/test_mori_transfer_engine_e2e.py @@ -130,7 +130,7 @@ class TestMoriTransferEngineE2E(PDDisaggregationServerBase): other_args=decode_args, ) - def test_generate_smoke(self): + def test_generate_basic(self): resp = requests.post( self.lb_url + "/generate", json={ @@ -273,7 +273,7 @@ class TestMoriTransferEngineTPMismatchE2E(PDDisaggregationServerBase): other_args=decode_args, ) - def test_generate_smoke_tp_mismatch(self): + def test_generate_with_tp_mismatch(self): resp = requests.post( self.lb_url + "/generate", json={ diff --git a/test/registered/debug_utils/comparator/test_entrypoint.py b/test/registered/debug_utils/comparator/test_entrypoint.py index a98ca6be1..eb7239e13 100644 --- a/test/registered/debug_utils/comparator/test_entrypoint.py +++ b/test/registered/debug_utils/comparator/test_entrypoint.py @@ -285,7 +285,7 @@ class TestEntrypointGroupingRaw: assert summary.total == 2 assert summary.passed == 2 - def test_text_output_smoke(self, tmp_path, capsys): + def test_text_output_format(self, tmp_path, capsys): """Text output format renders without errors and contains Config/Summary sections.""" baseline_path, target_path = _create_dumps(tmp_path, ["tensor_a"]) argv = _make_argv( diff --git a/test/registered/debug_utils/test_cuda_coredump_smoke.py b/test/registered/debug_utils/test_cuda_coredump.py similarity index 83% rename from test/registered/debug_utils/test_cuda_coredump_smoke.py rename to test/registered/debug_utils/test_cuda_coredump.py index 99ebb7f02..2354175fe 100644 --- a/test/registered/debug_utils/test_cuda_coredump_smoke.py +++ b/test/registered/debug_utils/test_cuda_coredump.py @@ -1,7 +1,7 @@ -"""Smoke test: intentionally trigger a CUDA illegal memory access +"""Intentionally trigger a CUDA illegal memory access to verify the coredump collection pipeline works end-to-end. -Manual use: python3 test/registered/debug_utils/test_cuda_coredump_smoke.py +Manual use: python3 test/registered/debug_utils/test_cuda_coredump.py """ import unittest @@ -17,7 +17,7 @@ register_cuda_ci( ) -class TestCudaCoredumpSmoke(unittest.TestCase): +class TestCudaCoredump(unittest.TestCase): def test_trigger_illegal_memory_access(self): x = torch.zeros(10, device="cuda") y = torch.arange(10, device="cuda") diff --git a/test/registered/hicache/test_hicache_storage_runtime_attach_detach.py b/test/registered/hicache/test_hicache_storage_runtime_attach_detach.py index 1517d6d4f..b46ca93f3 100644 --- a/test/registered/hicache/test_hicache_storage_runtime_attach_detach.py +++ b/test/registered/hicache/test_hicache_storage_runtime_attach_detach.py @@ -1,5 +1,5 @@ """ -E2E smoke test for HiCache storage runtime attach/detach. +E2E check for HiCache storage runtime attach/detach. This test launches an SGLang server with hierarchical cache enabled but WITHOUT any storage backend at startup, then attaches/detaches a storage backend via the diff --git a/test/registered/moe/test_cutedsl_moe.py b/test/registered/moe/test_cutedsl_moe.py index b0c10db79..d0ba9682d 100644 --- a/test/registered/moe/test_cutedsl_moe.py +++ b/test/registered/moe/test_cutedsl_moe.py @@ -581,7 +581,7 @@ class TestCuteDslV2(unittest.TestCase): Also checks both match the pure-PyTorch reference, and that a second cuda_graph pass reuses buffers deterministically (subsumes the former - cuda_graph_smoke test). + cuda_graph check). """ test_cases = [ # (num_tokens, hidden_size, intermediate_size, num_experts, top_k)