sgl-project/sglang

Pattern Findings

17209 matches across 20 categories. Click a row to expand file-level details.

Hyper-Verbose Identifiers9004 hits · 8835 pts

Severity	File	Line	Snippet
LOW	benchmark/dspy/bench_dspy_intro.py	136	def validate_context_and_answer(example, pred, trace=None):
LOW	…ing_window_attention_triton/bench_triton_swa_kernel.py	11	def extend_attention_fwd_torch(
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	122	def cleanup_flashinfer_workspace(ipc_handles):
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	85	def setup_flashinfer_workspace(
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	152	def get_trtllm_fused_allreduce_kwargs(self):
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	162	def flashinfer_fused_allreduce_rmsnorm(
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	287	def standard_allreduce_rmsnorm(
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	397	def standard_allreduce_rmsnorm_native(
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	478	def standard_allreduce_rmsnorm_native_compiled(
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	894	def prepare_results_with_speedups(results_dict):
LOW	…uler_batch/benchmark_write_req_to_token_pool_triton.py	12	def write_req_to_token_pool_triton(
LOW	…uler_batch/benchmark_write_req_to_token_pool_triton.py	49	def write_req_to_token_pool_triton_optimize(
LOW	…uler_batch/benchmark_write_req_to_token_pool_triton.py	91	def write_req_to_token_pool_reference(
LOW	…uler_batch/benchmark_write_req_to_token_pool_triton.py	114	def test_write_req_to_token_pool():
LOW	benchmark/kernels/fused_moe_triton/common_utils.py	23	def calculate_shard_intermediate_size(
LOW	benchmark/kernels/fused_moe_triton/common_utils.py	172	def get_rocm_configs_compute_bound() -> List[Dict[str, int]]:
LOW	benchmark/kernels/fused_moe_triton/common_utils.py	195	def get_configs_compute_bound() -> List[Dict[str, int]]:
LOW	…pseek/benchmark_deepgemm_dsv3_router_gemm_blackwell.py	137	def get_benchmark_plot_friendly(tp_sizes):
LOW	…nels/deepseek/benchmark_deepgemm_fp8_gemm_blackwell.py	231	def get_benchmark_plot_friendly(tp_size):
LOW	…hmark/kernels/quantization/tuning_block_wise_kernel.py	148	def get_rocm_configs_compute_bound():
LOW	…hmark/kernels/quantization/tuning_block_wise_kernel.py	171	def get_configs_compute_bound():
LOW	…s/decoding_attention_triton/triton_flashinfer_cudnn.py	98	def decode_attention_flashinfer(dtype, head_num_q, head_num_kv):
LOW	benchmark/mmmu/bench_sglang.py	127	async def process_sample_with_semaphore(
LOW	benchmark/mmmu/eval_utils.py	292	def _parse_explicit_multi_choice_answer(response, all_choices):
LOW	benchmark/mmmu/eval_utils.py	303	def parse_multi_choice_response(response, all_choices, index2ans):
LOW	…hmark/bench_pynccl_allocator/bench_segment_tracking.py	56	def bench_register_segments_with_comm(
LOW	…hmark/bench_pynccl_allocator/bench_segment_tracking.py	114	def bench_with_various_segment_counts(
LOW	benchmark/scheduler/bench_token_storage.py	84	def _batch_tensor_from_pyarrays(parts: list[array]) -> torch.Tensor:
LOW	benchmark/scheduler/bench_token_storage.py	256	def microbench_torch_tensor_paths(
LOW	benchmark/hicache/bench_mix.py	299	async def async_request_sglang_generate(
LOW	benchmark/hicache/bench_warm_cache.py	90	def _create_bench_client_session() -> aiohttp.ClientSession:
LOW	benchmark/hicache/bench_warm_cache.py	97	async def async_request_sglang_generate(
LOW	benchmark/hicache/bench_warm_cache.py	413	def maybe_write_summary_jsonl(
LOW	benchmark/hicache/bench_warm_cache.py	472	async def benchmark_shared_prefix_pct(
LOW	benchmark/hicache/bench_serving.py	71	async def async_request_openai_completions(
LOW	benchmark/hicache/data_processing.py	150	def sample_ultrachat_requests(
LOW	benchmark/hicache/data_processing.py	434	def sample_generated_shared_prefix_requests(
LOW	benchmark/prefill_only/bench_score.py	65	def create_score_request_builder():
LOW	benchmark/prefill_only/bench_score.py	80	def generate_text_with_token_count_local(num_toks):
LOW	benchmark/prefill_only/bench_score.py	120	def build_warmup_score_request() -> dict:
LOW	benchmark/prefill_only/util.py	46	def generate_text_with_token_count(
LOW	benchmark/prefill_only/util.py	106	def prepare_all_requests_parallel(
LOW	benchmark/prefill_only/util.py	398	async def perform_global_warmup_and_freeze(
LOW	benchmark/prefill_only/bench_embeddings.py	100	def validate_embeddings_response(response_data: dict) -> bool:
LOW	benchmark/prefill_only/bench_embeddings.py	110	def build_warmup_embeddings_request() -> dict:
LOW	…nchmark/bench_in_batch_prefix/bench_in_batch_prefix.py	76	def test_batch_by_batch_with_hint(all_prompts, gen_len):
LOW	benchmark/asr/bench_sglang.py	55	def run_asr_transcription_sync(client, model_name, y, sr, language=None):
LOW	benchmark/asr/bench_sglang.py	75	def run_asr_transcription_stream_sync(
LOW	benchmark/benchmark_batch/benchmark_tokenizer.py	179	def generate_random_token_ids(*, num_prompts, num_tokens, tokenizer):
LOW	benchmark/lora/lora_bench.py	48	async def async_request_openai_completions(
LOW	…ark/bench_linear_attention/bench_cutedsl_kda_decode.py	155	def run_prefill_then_decode_baseline(inp):
LOW	…ark/bench_linear_attention/bench_cutedsl_kda_decode.py	191	def run_prefill_then_decode_cutedsl(inp):
LOW	benchmark/generative_agents/agent_functions.py	56	def generate_event_triple_prompt(persona_name, action):
LOW	benchmark/generative_agents/agent_functions.py	89	def generate_pronunciatio_prompt(action):
LOW	benchmark/generative_agents/agent_functions.py	158	def action_location_sector_prompt(
LOW	benchmark/generative_agents/agent_functions.py	260	def action_location_object_prompt(
LOW	benchmark/llm_judge/bench_other.py	52	async def multi_dimension_judge_async(article, generate):
LOW	…l/sgl-router/tests/scripts/generate_parity_fixtures.py	63	def load_tokenizer_with_fallback(primary, fallback, slug):
LOW	experimental/sgl-router/tests/e2e/conftest.py	169	def build_smoke_router_config(
LOW	…sgl-router/tests/e2e/k8s_integration/test_lifecycle.py	54	def test_router_routes_after_scale_up(self, router_url):
8944 more matches not shown…

Decorative Section Separators1489 hits · 4868 pts

Severity	File	Line	Snippet
MEDIUM	…hmark/kernels/quantization/tuning_block_wise_kernel.py	13	# ==============================================================================
MEDIUM	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	128	# ---------------------------------------------------------------------------
MEDIUM	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	130	# ---------------------------------------------------------------------------
MEDIUM	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	181	# ---------------------------------------------------------------------------
MEDIUM	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	183	# ---------------------------------------------------------------------------
MEDIUM	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	333	# ---------------------------------------------------------------------------
MEDIUM	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	335	# ---------------------------------------------------------------------------
MEDIUM	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	390	# ---------------------------------------------------------------------------
MEDIUM	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	392	# ---------------------------------------------------------------------------
MEDIUM	benchmark/lora/lora_bench.py	13	# ==============================================================================
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	39	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	41	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	89	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	91	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	155	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	157	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	206	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	208	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	299	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_decode.py	301	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	38	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	40	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	68	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	70	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	117	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	119	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	173	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	175	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	232	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	234	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	323	# ---------------------------------------------------------------------------
MEDIUM	…rk/bench_linear_attention/bench_gdn_prefill_cutedsl.py	325	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	37	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	39	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	96	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	98	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	161	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	163	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	247	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	249	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	350	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	352	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	448	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_linear_attention/bench_gdn_prefill.py	450	# ---------------------------------------------------------------------------
MEDIUM	benchmark/bench_rope/benchmark_rope_index.py	21	# -----------------------------
MEDIUM	benchmark/bench_rope/benchmark_rope_index.py	23	# -----------------------------
MEDIUM	benchmark/bench_rope/benchmark_rope_index.py	39	# -----------------------------
MEDIUM	benchmark/bench_rope/benchmark_rope_index.py	41	# -----------------------------
MEDIUM	experimental/sgl-router/tests/e2e/conftest.py	246	# ---------------------------------------------------------------------------
MEDIUM	experimental/sgl-router/tests/e2e/conftest.py	248	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	37	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	39	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	49	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	51	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	74	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	76	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	81	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	83	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	88	# ---------------------------------------------------------------------------
MEDIUM	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	92	# ---------------------------------------------------------------------------
1429 more matches not shown…

Cross-File Repetition424 hits · 2120 pts

Severity	File	Line	Snippet
HIGH	…hmark/kernels/quantization/tuning_block_wise_kernel.py	0	this function performs matrix multiplication with block-wise quantization. it takes two input tensors `a` and `b` with s
HIGH	python/sglang/srt/layers/quantization/int8_kernel.py	0	this function performs matrix multiplication with block-wise quantization. it takes two input tensors `a` and `b` with s
HIGH	python/sglang/srt/layers/quantization/fp8_kernel.py	0	this function performs matrix multiplication with block-wise quantization. it takes two input tensors `a` and `b` with s
HIGH	benchmark/tree_of_thought_v0/bench_sglang.py	0	please generate a high-level plan for solving the following question. as the first step, just say what method and idea y
HIGH	benchmark/tree_of_thought_v0/bench_other.py	0	please generate a high-level plan for solving the following question. as the first step, just say what method and idea y
HIGH	benchmark/tree_of_thought_deep/bench_sglang.py	0	please generate a high-level plan for solving the following question. as the first step, just say what method and idea y
HIGH	benchmark/tree_of_thought_deep/lmql_funcs.py	0	please generate a high-level plan for solving the following question. as the first step, just say what method and idea y
HIGH	benchmark/tree_of_thought_deep/bench_other.py	0	please generate a high-level plan for solving the following question. as the first step, just say what method and idea y
HIGH	benchmark/tree_of_thought_v0/bench_sglang.py	0	the plan looks good! now, use real numbers and do the calculation. please solve the question step-by-step according to t
HIGH	benchmark/tree_of_thought_v0/bench_other.py	0	the plan looks good! now, use real numbers and do the calculation. please solve the question step-by-step according to t
HIGH	benchmark/tree_of_thought_deep/bench_sglang.py	0	the plan looks good! now, use real numbers and do the calculation. please solve the question step-by-step according to t
HIGH	benchmark/tree_of_thought_deep/lmql_funcs.py	0	the plan looks good! now, use real numbers and do the calculation. please solve the question step-by-step according to t
HIGH	benchmark/tree_of_thought_deep/bench_other.py	0	the plan looks good! now, use real numbers and do the calculation. please solve the question step-by-step according to t
HIGH	benchmark/tip_suggestion/bench_sglang.py	0	please expand a tip for a topic into a detailed paragraph. topic: staying healthy tip: regular exercise paragraph: incor
HIGH	benchmark/tip_suggestion/lmql_funcs.py	0	please expand a tip for a topic into a detailed paragraph. topic: staying healthy tip: regular exercise paragraph: incor
HIGH	benchmark/tip_suggestion/bench_other.py	0	please expand a tip for a topic into a detailed paragraph. topic: staying healthy tip: regular exercise paragraph: incor
HIGH	benchmark/json_jump_forward/bench_sglang.py	0	"house": "(gryffindor\|slytherin\|ravenclaw\|hufflepuff)",\n
HIGH	benchmark/json_jump_forward/bench_other.py	0	"house": "(gryffindor\|slytherin\|ravenclaw\|hufflepuff)",\n
HIGH	examples/frontend_language/usage/json_decode.py	0	"house": "(gryffindor\|slytherin\|ravenclaw\|hufflepuff)",\n
HIGH	benchmark/json_jump_forward/bench_sglang.py	0	"blood status": "(pure-blood\|half-blood\|muggle-born)",\n
HIGH	benchmark/json_jump_forward/bench_other.py	0	"blood status": "(pure-blood\|half-blood\|muggle-born)",\n
HIGH	examples/frontend_language/usage/json_decode.py	0	"blood status": "(pure-blood\|half-blood\|muggle-born)",\n
HIGH	benchmark/json_jump_forward/bench_sglang.py	0	"occupation": "(student\|teacher\|auror\|ministry of magic\|death eater\|order of the phoenix)",\n
HIGH	benchmark/json_jump_forward/bench_other.py	0	"occupation": "(student\|teacher\|auror\|ministry of magic\|death eater\|order of the phoenix)",\n
HIGH	examples/frontend_language/usage/json_decode.py	0	"occupation": "(student\|teacher\|auror\|ministry of magic\|death eater\|order of the phoenix)",\n
HIGH	benchmark/tree_of_thought_deep/bench_sglang.py	0	okay. now, evaluate your own solution and give it a score on a scale of 1 to 5. please do rigorous check of the correctn
HIGH	benchmark/tree_of_thought_deep/lmql_funcs.py	0	okay. now, evaluate your own solution and give it a score on a scale of 1 to 5. please do rigorous check of the correctn
HIGH	benchmark/tree_of_thought_deep/bench_other.py	0	okay. now, evaluate your own solution and give it a score on a scale of 1 to 5. please do rigorous check of the correctn
HIGH	benchmark/tree_of_thought_deep/bench_sglang.py	0	based on your reflection, do you change your mind? now, give me the final answer after careful consideration.
HIGH	benchmark/tree_of_thought_deep/lmql_funcs.py	0	based on your reflection, do you change your mind? now, give me the final answer after careful consideration.
HIGH	benchmark/tree_of_thought_deep/bench_other.py	0	based on your reflection, do you change your mind? now, give me the final answer after careful consideration.
HIGH	test/srt/cpu/test_qwen3.py	0	derives `query`, `key` and `value` tensors from `mixed_qkvzba`.
HIGH	test/registered/cpu/test_qwen3.py	0	derives `query`, `key` and `value` tensors from `mixed_qkvzba`.
HIGH	python/sglang/srt/models/qwen3_5.py	0	derives `query`, `key` and `value` tensors from `mixed_qkvzba`.
HIGH	python/sglang/srt/models/qwen3_next.py	0	derives `query`, `key` and `value` tensors from `mixed_qkvzba`.
HIGH	test/srt/cpu/utils.py	0	matrix multiplication function that supports per-token input quantization and per-column weight quantization
HIGH	test/registered/quant/test_int8_kernel.py	0	matrix multiplication function that supports per-token input quantization and per-column weight quantization
HIGH	test/registered/cpu/utils.py	0	matrix multiplication function that supports per-token input quantization and per-column weight quantization
HIGH	…t/registered/moe/test_triton_moe_channel_fp8_kernel.py	0	matrix multiplication function that supports per-token input quantization and per-column weight quantization
HIGH	test/srt/cpu/utils.py	0	this function performs fused moe with per-column int8 quantization using native torch.
HIGH	test/registered/quant/test_int8_kernel.py	0	this function performs fused moe with per-column int8 quantization using native torch.
HIGH	test/registered/cpu/utils.py	0	this function performs fused moe with per-column int8 quantization using native torch.
HIGH	…t/registered/moe/test_triton_moe_channel_fp8_kernel.py	0	this function performs fused moe with per-column int8 quantization using native torch.
HIGH	test/srt/cpu/test_causal_conv1d.py	0	x: (batch, dim, seqlen) weight: (dim, width) bias: (dim,) initial_states: (batch, dim, width - 1) final_states_out: (bat
HIGH	test/registered/layers/mamba/test_causal_conv1d.py	0	x: (batch, dim, seqlen) weight: (dim, width) bias: (dim,) initial_states: (batch, dim, width - 1) final_states_out: (bat
HIGH	test/registered/cpu/test_causal_conv1d.py	0	x: (batch, dim, seqlen) weight: (dim, width) bias: (dim,) initial_states: (batch, dim, width - 1) final_states_out: (bat
HIGH	sgl-kernel/tests/test_causal_conv1d.py	0	x: (batch, dim, seqlen) weight: (dim, width) bias: (dim,) initial_states: (batch, dim, width - 1) final_states_out: (bat
HIGH	test/srt/cpu/test_causal_conv1d.py	0	x: (batch, dim) or (batch, dim, seqlen) conv_state: (batch, dim, state_len), where state_len >= width - 1 weight: (dim,
HIGH	test/registered/layers/mamba/test_causal_conv1d.py	0	x: (batch, dim) or (batch, dim, seqlen) conv_state: (batch, dim, state_len), where state_len >= width - 1 weight: (dim,
HIGH	test/registered/cpu/test_causal_conv1d.py	0	x: (batch, dim) or (batch, dim, seqlen) conv_state: (batch, dim, state_len), where state_len >= width - 1 weight: (dim,
HIGH	sgl-kernel/tests/test_causal_conv1d.py	0	x: (batch, dim) or (batch, dim, seqlen) conv_state: (batch, dim, state_len), where state_len >= width - 1 weight: (dim,
HIGH	test/manual/quant/test_block_fp8.py	0	this function performs matrix multiplication with block-wise quantization using native torch. it takes two input tensors
HIGH	test/manual/quant/test_block_fp8_deep_gemm_blackwell.py	0	this function performs matrix multiplication with block-wise quantization using native torch. it takes two input tensors
HIGH	test/registered/quant/test_block_int8.py	0	this function performs matrix multiplication with block-wise quantization using native torch. it takes two input tensors
HIGH	test/manual/nightly/test_vlms_vit_cuda_graph.py	0	evaluate a vlm on the mmmu validation set with lmms‑eval. only `model_version` (checkpoint) and `chat_template` vary; we
HIGH	test/manual/nightly/test_vlms_vit_flashinfer_cudnn.py	0	evaluate a vlm on the mmmu validation set with lmms‑eval. only `model_version` (checkpoint) and `chat_template` vary; we
HIGH	test/manual/nightly/test_vlms_piecewise_cuda_graph.py	0	evaluate a vlm on the mmmu validation set with lmms‑eval. only `model_version` (checkpoint) and `chat_template` vary; we
HIGH	python/sglang/test/ascend/vlm_utils.py	0	evaluate a vlm on the mmmu validation set with lmms‑eval. only `model_version` (checkpoint) and `chat_template` vary; we
HIGH	test/manual/nightly/test_vlms_vit_cuda_graph.py	0	common method to run vlm mmmu benchmark test. args: model: model to test output_path: path for output logs test_name: op
HIGH	test/manual/nightly/test_vlms_vit_flashinfer_cudnn.py	0	common method to run vlm mmmu benchmark test. args: model: model to test output_path: path for output logs test_name: op
364 more matches not shown…

Unused Imports1818 hits · 1706 pts

Severity	File	Line
LOW	benchmark/scheduler/bench_token_storage.py	25
LOW	benchmark/hicache/perf.py	1
LOW	…router/tests/scripts/generate_kv_events_hash_parity.py	48
LOW	…perimental/sgl-router/tests/e2e/test_tokenize_smoke.py	5
LOW	experimental/sgl-router/tests/e2e/conftest.py	21
LOW	experimental/sgl-router/tests/e2e/test_chat_smoke.py	5
LOW	experimental/sgl-router/tests/e2e/test_chat_smoke.py	8
LOW	experimental/sgl-router/tests/e2e/infra/gateway.py	26
LOW	experimental/sgl-router/tests/e2e/infra/model_specs.py	16
LOW	experimental/sgl-router/tests/e2e/infra/model_pool.py	19
LOW	…al/sgl-router/tests/e2e/k8s_integration/fake_worker.py	10
LOW	…sgl-router/tests/e2e/k8s_integration/test_lifecycle.py	14
LOW	…sgl-router/tests/e2e/k8s_integration/test_lifecycle.py	16
LOW	…sgl-router/tests/e2e/k8s_integration/test_lifecycle.py	19
LOW	…sgl-router/tests/e2e/k8s_integration/test_lifecycle.py	20
LOW	…ental/sgl-router/tests/e2e/k8s_integration/conftest.py	12
LOW	…sgl-router/tests/e2e/k8s_integration/test_discovery.py	8
LOW	…sgl-router/tests/e2e/k8s_integration/test_discovery.py	11
LOW	…sgl-router/tests/e2e/k8s_integration/test_discovery.py	12
LOW	…uter/tests/e2e/k8s_integration/test_cross_namespace.py	15
LOW	…uter/tests/e2e/k8s_integration/test_cross_namespace.py	18
LOW	…outer/tests/e2e/k8s_integration/test_reconciliation.py	16
LOW	…outer/tests/e2e/k8s_integration/test_reconciliation.py	18
LOW	…l-router/tests/e2e/chat_completions/test_validation.py	14
LOW	…ts/e2e/chat_completions/test_two_router_convergence.py	38
LOW	test/manual/test_ray_engine.py	25
LOW	test/manual/quant/test_block_fp8.py	679
LOW	test/manual/layers/moe/bench_mxfp4_sm90_kernels.py	21
LOW	test/manual/hicache/test_pp_with_hicache.py	79
LOW	…ion/unittests/dual_chunk/test_dual_chunk_flash_attn.py	48
LOW	…ed/unit/layers/quantization/test_mxfp4_sm90_cutlass.py	14
LOW	…ed/unit/layers/quantization/test_mxfp4_sm90_cutlass.py	174
LOW	…d/unit/hardware_backend/mlx/test_attention_patching.py	3
LOW	…istered/unit/hardware_backend/mlx/test_quantization.py	13
LOW	test/registered/unit/distributed/test_parallel_state.py	37
LOW	…egistered/unit/utils/test_hf_transformers_fastokens.py	19
LOW	…t/registered/unit/model_loader/test_modelopt_export.py	29
LOW	…saggregation/test_disaggregation_decode_radix_cache.py	24
LOW	…saggregation/test_disaggregation_decode_radix_cache.py	32
LOW	test/registered/layers/test_fla_layernorm_guard.py	1
LOW	…gistered/models_e2e/test_deepseek_v4_flash_fp4_h200.py	28
LOW	…st/registered/openai_server/basic/test_http2_server.py	24
LOW	test/registered/prefill_only/test_serving_rerank.py	134
LOW	test/registered/observability/test_tracing.py	57
LOW	test/registered/vlm/test_vision_openai_server_a.py	12
LOW	test/registered/amd/test_wan22_fp8_mla.py	3
LOW	…t/registered/debug_utils/comparator/testing_helpers.py	3
LOW	…gistered/debug_utils/comparator/test_meta_overrider.py	3
LOW	test/registered/debug_utils/comparator/test_e2e_demo.py	9
LOW	…tils/comparator/aligner/token_aligner/test_executor.py	1
LOW	…ered/breakable_cuda_graph/test_breakable_cuda_graph.py	37
LOW	…ered/breakable_cuda_graph/test_breakable_cuda_graph.py	52
LOW	…ered/breakable_cuda_graph/test_breakable_cuda_graph.py	198
LOW	…ered/breakable_cuda_graph/test_breakable_cuda_graph.py	260
LOW	python/sglang/kernel_api_logging.py	7
LOW	python/sglang/__init__.py	35
LOW	python/sglang/__init__.py	36
LOW	python/sglang/__init__.py	36
LOW	python/sglang/__init__.py	36
LOW	python/sglang/__init__.py	36
1758 more matches not shown…

Excessive Try-Catch Wrapping1416 hits · 1670 pts

Severity	File	Line	Snippet
LOW	benchmark/bench_adaptive_speculative.py	70	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	117	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	130	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	47	except Exception: # pragma: no cover - fallback on non-supported platforms
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	650	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	664	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	683	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	700	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	720	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	737	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	762	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	787	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	811	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	828	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	854	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	882	except Exception as e:
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	1083	except Exception as e:
LOW	benchmark/kernels/all_reduce/benchmark_aiter.py	227	except Exception as e:
LOW	benchmark/kernels/all_reduce/benchmark_aiter.py	237	except Exception as e:
LOW	benchmark/kernels/all_reduce/benchmark_aiter.py	253	except Exception as e:
LOW	benchmark/kernels/all_reduce/benchmark_aiter.py	265	except Exception as e:
LOW	benchmark/kernels/all_reduce/benchmark_aiter.py	305	except Exception:
LOW	benchmark/kernels/all_reduce/benchmark_all_reduce.py	245	except Exception as e:
LOW	benchmark/kernels/all_reduce/benchmark_all_reduce.py	255	except Exception as e:
LOW	benchmark/kernels/all_reduce/benchmark_all_reduce.py	274	except Exception as e:
LOW	benchmark/kernels/all_reduce/benchmark_all_reduce.py	284	except Exception as e:
LOW	benchmark/kernels/all_reduce/benchmark_all_reduce.py	324	except Exception:
LOW	benchmark/kernels/quantization/bench_fp4_quant.py	114	except Exception as e:
LOW	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	228	except Exception:
LOW	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	305	except Exception:
LOW	benchmark/mmmu/bench_sglang.py	64	except Exception:
LOW	benchmark/mmmu/bench_hf.py	35	except Exception as first_exception:
LOW	benchmark/mmmu/bench_hf.py	60	except Exception as second_exception:
LOW	benchmark/mmmu/eval_utils.py	195	except Exception as exc:
MEDIUM	benchmark/mmmu/eval_utils.py	514	print(f"Error to extract answer from: {pred_i}")
LOW	benchmark/json_schema/bench_sglang.py	104	except Exception as e:
MEDIUM	…hmark/bench_pynccl_allocator/bench_segment_tracking.py	195	print("Error: CUDA is not available. This benchmark requires a GPU.")
LOW	benchmark/hicache/bench_mix.py	362	except Exception as e:
LOW	benchmark/hicache/bench_mix.py	423	except Exception as e:
MEDIUM	benchmark/hicache/bench_mix.py	479	print(f"Error processing response for client {user_data}: {e}")
MEDIUM	benchmark/hicache/bench_mix.py	417	def handle_request(self, user_data):
LOW	benchmark/hicache/bench_warm_cache.py	169	except Exception as exc:
LOW	benchmark/hicache/bench_serving.py	194	except Exception:
LOW	benchmark/hicache/bench_serving.py	214	except Exception:
LOW	benchmark/hicache/bench_serving.py	273	except Exception as e:
MEDIUM	benchmark/hicache/bench_serving.py	632	print(f"Error running benchmark for request rate: {request_rate}")
LOW	benchmark/hicache/bench_serving.py	739	except Exception as e:
MEDIUM	benchmark/hicache/bench_multiturn.py	180	print(f"Error writing to JSONL file: {e}")
LOW	benchmark/hicache/bench_multiturn.py	395	except Exception as e:
MEDIUM	benchmark/hicache/bench_multiturn.py	543	print(f"Error processing response for client {client_id}: {e}")
LOW	benchmark/hicache/bench_multiturn.py	552	except Exception as e:
LOW	benchmark/prefill_only/bench_score.py	108	except Exception as e:
MEDIUM	benchmark/prefill_only/bench_score.py	109	print(f"Error building request {index}: {e}")
LOW	benchmark/prefill_only/util.py	131	except Exception as e:
MEDIUM	benchmark/prefill_only/util.py	132	print(f"Error building request {index}: {e}")
LOW	benchmark/prefill_only/util.py	160	except Exception as e:
MEDIUM	benchmark/prefill_only/util.py	161	print(f"Error processing request result: {e}")
LOW	benchmark/prefill_only/util.py	278	except Exception as e:
LOW	benchmark/prefill_only/util.py	327	except Exception as e:
MEDIUM	benchmark/prefill_only/util.py	328	print(f"Error sending {profile_text} request: {e}")
1356 more matches not shown…

Deep Nesting1463 hits · 1308 pts

Severity	File	Line
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	605
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	894
LOW	…hmark/kernels/all_reduce/benchmark_fused_ar_rms_amd.py	367
LOW	…/fused_moe_triton/benchmark_torch_compile_fused_moe.py	16
LOW	benchmark/kernels/fused_moe_triton/common_utils.py	32
LOW	benchmark/kernels/fused_moe_triton/common_utils.py	172
LOW	benchmark/kernels/fused_moe_triton/common_utils.py	195
LOW	…hmark/kernels/quantization/tuning_block_wise_kernel.py	148
LOW	…hmark/kernels/quantization/tuning_block_wise_kernel.py	171
LOW	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	133
LOW	benchmark/kernels/lora_csgmv/tune_lora_csgmv.py	151
LOW	benchmark/kernels/deepep/deepep_utils.py	157
LOW	benchmark/kernels/deepep/tuning_deepep.py	30
LOW	…s/decoding_attention_triton/triton_flashinfer_cudnn.py	159
LOW	benchmark/mmmu/eval_utils.py	154
LOW	benchmark/mmmu/eval_utils.py	303
LOW	benchmark/mmmu/eval_utils.py	423
LOW	benchmark/mmmu/eval_utils.py	528
LOW	benchmark/mmmu/eval_utils.py	430
LOW	benchmark/json_schema/bench_sglang.py	29
LOW	benchmark/line_retrieval/bench_sglang.py	30
LOW	benchmark/hicache/bench_mix.py	299
LOW	benchmark/hicache/nextqa.py	9
LOW	benchmark/hicache/bench_warm_cache.py	97
LOW	benchmark/hicache/bench_serving.py	71
LOW	benchmark/hicache/bench_serving.py	286
LOW	benchmark/hicache/data_processing.py	37
LOW	benchmark/hicache/data_processing.py	198
LOW	benchmark/hicache/data_processing.py	267
LOW	benchmark/hicache/data_processing.py	520
LOW	benchmark/hicache/bench_multiturn.py	435
LOW	benchmark/hicache/bench_multiturn.py	555
LOW	benchmark/prefill_only/util.py	106
LOW	benchmark/prefill_only/util.py	284
LOW	benchmark/reasoning_benchmark/answer_extraction.py	8
LOW	benchmark/reasoning_benchmark/answer_extraction.py	182
LOW	benchmark/reasoning_benchmark/answer_extraction.py	214
LOW	benchmark/reasoning_benchmark/eval_utils.py	59
LOW	benchmark/asr/bench_sglang.py	75
LOW	benchmark/json_jump_forward/bench_other.py	132
LOW	benchmark/lora/lora_bench.py	48
LOW	benchmark/bench_linear_attention/bench_gdn_decode.py	359
LOW	benchmark/bench_linear_attention/bench_gdn_prefill.py	520
LOW	experimental/sgl-router/tests/e2e/infra/gateway.py	237
LOW	experimental/sgl-router/tests/e2e/infra/gateway.py	378
LOW	experimental/sgl-router/tests/e2e/infra/model_pool.py	106
LOW	…outer/tests/e2e/k8s_integration/test_reconciliation.py	85
LOW	…l-router/tests/e2e/chat_completions/test_validation.py	62
LOW	test/srt/cpu/test_decode.py	264
LOW	test/srt/cpu/test_topk.py	119
LOW	test/manual/test_quick_allreduce.py	106
LOW	test/manual/test_quick_allreduce.py	274
LOW	test/manual/test_triton_attention_rocm_mla.py	224
LOW	test/manual/test_logprobs.py	99
LOW	test/manual/test_logprobs.py	302
LOW	test/manual/test_triton_moe_wna16.py	16
LOW	test/manual/test_mscclpp.py	106
LOW	test/manual/test_kv_events.py	28
LOW	test/manual/test_kv_events.py	164
LOW	test/manual/test_kv_events.py	292
1403 more matches not shown…

Over-Commented Block795 hits · 777 pts

Severity	File	Line	Snippet
LOW	…hmark/kernels/quantization/tuning_block_wise_kernel.py	1	# Copyright 2025 SGLang Team
LOW	benchmark/lora/lora_bench.py	1	# Copyright 2023-2024 SGLang Team
LOW	…ental/sgl-router/tests/proxy/pd_bootstrap_injection.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	…perimental/sgl-router/tests/proxy/graceful_shutdown.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	…perimental/sgl-router/tests/proxy/pd_pool_isolation.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	experimental/sgl-router/tests/proxy/chat_routing.rs	701	.await;
LOW	experimental/sgl-router/tests/proxy/chat_routing.rs	1201	// Wait long enough for the SSE pump to notice the receiver-drop and
LOW	experimental/sgl-router/tests/proxy/common/streaming.rs	21	use http_body_util::BodyExt;
LOW	…imental/sgl-router/tests/component/tokenizer/parity.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	…sgl-router/tests/component/policies/cache_aware_zmq.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	…/tests/component/policies/kv_events_two_subscribers.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	…/tests/component/policies/kv_events_two_subscribers.rs	141	"router_b shutdown after router_a drained took {elapsed:?}; \
LOW	…/tests/component/policies/kv_events_tree_concurrent.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	…uter/tests/component/policies/kv_events_hash_parity.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	…sgl-router/tests/component/workers/concurrent_state.rs	81	let snapshot = r.workers_for(&model);
LOW	experimental/sgl-router/tests/e2e/conftest.py	41	# Mirrors SMG's e2e_test/conftest.py sys.path setup.
LOW	experimental/sgl-router/tests/e2e/requirements.txt	1	httpx==0.27.2
LOW	experimental/sgl-router/benches/tree_lookup.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	experimental/sgl-router/src/proxy/sse.rs	21	///
LOW	experimental/sgl-router/src/proxy/sse.rs	41	/// The opaque `Box<dyn Send + 'static>` accepts any drop-only payload — most
LOW	experimental/sgl-router/src/config/types.rs	61	stale_request_timeout_secs: default_stale_request_timeout_secs(),
LOW	experimental/sgl-router/src/config/types.rs	141
LOW	experimental/sgl-router/src/config/types.rs	201	}
LOW	experimental/sgl-router/src/config/types.rs	321	/// Two operating modes, distinguished by which selector fields are set:
LOW	experimental/sgl-router/src/config/types.rs	341	/// `WorkerMode` and `bootstrap_port` for each worker are filled in by
LOW	experimental/sgl-router/src/config/types.rs	361	///
LOW	experimental/sgl-router/src/config/types.rs	421	}
LOW	experimental/sgl-router/src/config/types.rs	501	// watcher (`watcher::Config::default().labels(&label)`
LOW	experimental/sgl-router/src/tokenizer/mod.rs	101	// tiny BPE fixture is byte-level and lossless for ASCII.
LOW	experimental/sgl-router/src/tokenizer/mod.rs	141	/// registry's `Arc<Tokenizer>` is `Send + Sync` and that
LOW	experimental/sgl-router/src/health/circuit_breaker.rs	61	config,
LOW	experimental/sgl-router/src/discovery/types.rs	41	/// new worker becomes available, and [`DiscoveryEvent::Removed`] when it
LOW	experimental/sgl-router/src/discovery/types.rs	61	}
LOW	experimental/sgl-router/src/discovery/k8s.rs	81	///
LOW	experimental/sgl-router/src/discovery/k8s.rs	141	});
LOW	experimental/sgl-router/src/discovery/k8s.rs	221	Ok(())
LOW	experimental/sgl-router/src/discovery/k8s.rs	741
LOW	experimental/sgl-router/src/discovery/static_urls.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	experimental/sgl-router/src/server/error.rs	21	/// build error). `source` captures the full anyhow chain for server-side
LOW	experimental/sgl-router/src/server/error.rs	41	UpstreamStatus { status: StatusCode },
LOW	experimental/sgl-router/src/server/error.rs	61	/// independently of full-model outages.
LOW	experimental/sgl-router/src/server/error.rs	81	#[error("stale request expired for model {model}")]
LOW	experimental/sgl-router/src/server/metrics.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	experimental/sgl-router/src/server/metrics.rs	21	//! \| `sgl_router_requests_total` \| Counter \| `worker_url`, `model_id`, `mode`, `outcome` \|
LOW	experimental/sgl-router/src/server/routes/chat.rs	21	/// affinity for a PD-disaggregated request. The router fans the
LOW	experimental/sgl-router/src/server/routes/chat.rs	41	/// for normal chat traffic (a 200 k-token context tokenized as JSON is well
LOW	experimental/sgl-router/src/server/routes/chat.rs	181	// ends, the client disconnects, or the handler returns an error. In
LOW	experimental/sgl-router/src/server/routes/chat.rs	201	// Snapshot the labels we need for metrics BEFORE moving the worker
LOW	experimental/sgl-router/src/server/routes/chat.rs	221	//
LOW	experimental/sgl-router/src/server/routes/chat.rs	341	);
LOW	experimental/sgl-router/src/server/routes/chat.rs	421	}
LOW	experimental/sgl-router/src/server/routes/chat.rs	441	(body.len() / CHARS_PER_TOKEN_ESTIMATE).max(1)
LOW	experimental/sgl-router/src/server/routes/chat.rs	461	/// to this address for the KV transfer.
LOW	experimental/sgl-router/src/server/routes/chat.rs	501	let bytes = serde_json::to_vec(&obj).map_err(\|e\| {
LOW	experimental/sgl-router/src/server/routes/chat.rs	641	}
LOW	experimental/sgl-router/src/policies/registry.rs	1	// SPDX-FileCopyrightText: Copyright (c) 2026 The SGLang Authors
LOW	experimental/sgl-router/src/policies/registry.rs	21	//!
LOW	experimental/sgl-router/src/policies/registry.rs	41	/// decode peer is considered "too hot" — we fall back to the lowest-load
LOW	experimental/sgl-router/src/policies/registry.rs	61	/// empty pool as a transient failure, not as "zero work".
LOW	experimental/sgl-router/src/policies/registry.rs	101	/// Returns `Err(NoHealthyWorkers)` only when the model has zero
735 more matches not shown…

Self-Referential Comments142 hits · 432 pts

Severity	File	Line	Snippet
MEDIUM	benchmark/dspy/bench_dspy_intro.py	95	# Define the predictor.
MEDIUM	benchmark/dspy/bench_dspy_intro.py	107	# Define the predictor. Notice we're just changing the class. The signature BasicQA is unchanged.
MEDIUM	benchmark/mmmu/data_utils.py	157	# Create a JSON object with the filename as the key and caption as the value
MEDIUM	benchmark/hicache/bench_mix.py	24	# Create a lock for thread-safe debug log writing
MEDIUM	benchmark/prefill_only/bench_score.py	157	# Create the request builder function with shared tokenizer
MEDIUM	benchmark/prefill_only/util.py	173	# Create the full request list by cycling through unique requests
MEDIUM	benchmark/multi_turn_chat/long_prompt_multi_turn.py	29	# Create a unique cache filename based on the arguments that affect generation
MEDIUM	experimental/sgl-router/tests/e2e/infra/gateway.py	45	# This file is at `experimental/sgl-router/tests/e2e/infra/gateway.py`,
MEDIUM	test/srt/cpu/test_mamba.py	14	"""This function is intended to align with the l2norm implementation in the FLA library."""
MEDIUM	test/manual/test_weight_validation.py	75	# Create a minimal valid safetensors file
MEDIUM	test/manual/test_weight_validation.py	166	# Create a broken symlink for the index file
MEDIUM	test/manual/test_modelopt_fp8kvcache.py	15	# Create a ModelOptFp8Config object
MEDIUM	test/manual/test_config_integration.py	27	# Create a temporary config file
MEDIUM	test/manual/test_config_integration.py	82	# Create a temporary config file
MEDIUM	test/manual/test_config_integration.py	113	# Create a temporary config file
MEDIUM	test/manual/test_async_dynamic_batch_tokenizer.py	54	# Create a proper BatchEncoding-like object that supports dict operations
MEDIUM	test/manual/test_async_dynamic_batch_tokenizer.py	193	# Create a new async tokenizer with a failing tokenizer
MEDIUM	test/manual/attention/test_flashattn_backend.py	66	# Create a large enough req_to_token_pool to fit the test usage.
MEDIUM	test/manual/attention/test_flashattn_mla_backend.py	54	# Create a proper req_to_token_pool with the req_to_token attribute
MEDIUM	…ual/lang_frontend/test_separate_reasoning_execution.py	72	# Create a mock backend to avoid AttributeError in __del__
MEDIUM	…ual/lang_frontend/test_separate_reasoning_execution.py	75	# Create a StreamExecutor with necessary setup
MEDIUM	…ual/lang_frontend/test_separate_reasoning_execution.py	106	# Create a gen expression and a separate_reasoning expression
MEDIUM	test/manual/debug_utils/run_with_retry.py	37	# Create a TestFile with a reasonable estimated time
MEDIUM	test/manual/eval/test_longbench_v2_eval.py	73	# Create a temporary JSON file with sample data
MEDIUM	test/registered/quant/test_quant_config_parsing.py	23	# Create a raw instance
MEDIUM	test/registered/kernels/test_dsa_indexer.py	78	# Create a simple page table for testing
MEDIUM	test/registered/kernels/test_dsa_indexer.py	94	# Create a simple page table for testing with page size 1
MEDIUM	…st/registered/attention/test_wave_attention_kernels.py	186	# Define the varying parameter values
MEDIUM	…/registered/attention/test_triton_attention_kernels.py	314	# Define the varying parameter values
MEDIUM	…egistered/attention/test_normal_decode_set_metadata.py	128	# Create a simple SWA KV pool for testing
MEDIUM	…egistered/attention/test_normal_decode_set_metadata.py	149	# Create a minimal mock that inherits from SWAKVPool to pass isinstance check
MEDIUM	test/registered/unit/distributed/test_parallel_state.py	95	# Create a mock group object
MEDIUM	test/registered/unit/distributed/test_parallel_state.py	195	# Create a mock group object
MEDIUM	…stered/unit/function_call/test_function_call_parser.py	517	# Create a concrete implementation of BaseFormatDetector for testing
MEDIUM	…tered/unit/mem_cache/test_radix_cache_slru_accuracy.py	56	# Create a cache with the memory pools
MEDIUM	…ed/unit/mem_cache/test_unified_radix_cache_unittest.py	1295	# Create a chain: root -> A -> B -> C (3 levels)
MEDIUM	…t/registered/unit/model_loader/test_modelopt_loader.py	72	# Create a basic model config with unified quantization flag
MEDIUM	…t/registered/unit/model_loader/test_modelopt_loader.py	116	# Create a custom load_model method for testing that simulates the real logic
MEDIUM	…egistered/unit/entrypoints/openai/test_serving_chat.py	524	# Create a mock conversation object that will be returned by generate_chat_conv
MEDIUM	…egistered/unit/entrypoints/openai/test_serving_chat.py	1211	# Create a mock conversation object
MEDIUM	test/registered/cpu/utils.py	465	# Create a new tensor with alternating values
MEDIUM	test/registered/cpu/utils.py	488	# Create a tensor with the E2M1 values
MEDIUM	test/registered/cpu/test_mamba.py	17	"""This function is intended to align with the l2norm implementation in the FLA library."""
MEDIUM	…registered/hicache/test_hicache_storage_3fs_backend.py	26	# Create a temporary JSON config file for HF3FS
MEDIUM	test/registered/lora/test_chunked_sgmv_backend.py	189	# Create a minimal mock ForwardBatch for the test
MEDIUM	test/registered/lora/test_chunked_sgmv_backend.py	221	# Create a minimal backend instance to access _get_segments_info
MEDIUM	test/registered/rl/test_update_weights_from_tensor.py	136	# Create a small set of parameters for testing
MEDIUM	test/registered/rl/test_update_weights_from_tensor.py	150	# Create a flattened bucket
MEDIUM	test/registered/rl/test_update_weights_from_tensor.py	157	# Create the dict format expected by _update_weights_from_flattened_bucket
MEDIUM	test/registered/rl/test_update_weights_from_tensor.py	167	# Create a list where each rank contains the same serialized data
MEDIUM	python/sglang/bench_serving.py	1270	# Create a temporary DatasetRow object for warmup
MEDIUM	python/sglang/bench_serving.py	1286	# Create the test input once
MEDIUM	python/sglang/bench_serving.py	2309	# Create a mutually exclusive group for profiling URLs
MEDIUM	python/sglang/test/nightly_bench_utils.py	72	# Create a combined link or use the first available one
MEDIUM	python/sglang/jit_kernel/ngram_embedding.py	93	# Create an empty tensor for ignore_tokens
MEDIUM	python/sglang/srt/layers/attention/flashmla_backend.py	524	# Create a dummy forward_mode for draft step
MEDIUM	…/sglang/srt/layers/attention/flashattention_backend.py	2477	# Create a modified version for local attention that only processes the last token
MEDIUM	…/sglang/srt/layers/attention/flashattention_backend.py	3262	# Create a block_table for the local attention blocks
MEDIUM	…/sglang/srt/layers/attention/dsa/index_buf_accessor.py	760	# Define the token range within the block and the K dimension range handled by the thread.
MEDIUM	python/sglang/srt/layers/attention/fla/kda.py	4	# This file contains code copied from the flash-linear-attention project.
82 more matches not shown…

Redundant / Tautological Comments266 hits · 405 pts

Severity	File	Line	Snippet
LOW	…shinfer_allreduce_fusion/benchmark_fused_collective.py	1153	# Check if running with torchrun (required for collective operations)
LOW	benchmark/json_schema/bench_sglang.py	97	# Check if the outputs are valid
LOW	benchmark/ceval/bench_sglang.py	112	# Print results
LOW	benchmark/prefill_only/util.py	163	# Check if we have any valid requests
LOW	benchmark/boolq/bench_sglang.py	89	# Print results
LOW	benchmark/gsm8k/bench_sglang.py	143	# Print results
LOW	benchmark/gsm8k/bench_other.py	129	# Print results
LOW	benchmark/mmlu/bench_sglang.py	179	# Print results
LOW	test/manual/test_schedule_policy.py	76	# Check if FCFS keeps the original order
LOW	test/manual/test_schedule_policy.py	101	# Check if priority enabled fcfs ordering is applied.
LOW	test/manual/test_schedule_policy.py	127	# Check if priority enabled fcfs ordering is applied.
LOW	test/manual/test_schedule_policy.py	149	# Check if priority enabled fcfs ordering is applied.
LOW	test/manual/test_schedule_policy.py	171	# Check if priority enabled fcfs ordering is applied.
LOW	test/manual/test_schedule_policy.py	195	# Check if priority enabled fcfs ordering is applied.
LOW	…nual/kv_transfer/test_mooncake_transfer_engine_init.py	201	# Check if mooncake should be used
LOW	…tered/unit/mem_cache/test_radix_cache_slru_accuracy.py	108	# Check if the frequently accessed key-value is still present
LOW	…tered/unit/mem_cache/test_radix_cache_slru_accuracy.py	114	# Check if the first low-frequency key-value has been evicted
LOW	…tered/unit/mem_cache/test_radix_cache_slru_accuracy.py	126	# Check if the tensor is empty, which indicates the key was not found (evicted)
LOW	…t/registered/unit/model_loader/test_modelopt_export.py	27	# Check if modelopt is available
LOW	…d/unit/batch_invariant_ops/test_batch_invariant_ops.py	53	# Check if results are identical
LOW	…d/unit/batch_invariant_ops/test_batch_invariant_ops.py	193	# Check if results are identical
LOW	…registered/disaggregation/test_disaggregation_basic.py	206	# Check if servers are still healthy
LOW	…/ascend/interface/test_npu_openai_server_ignore_eos.py	84	# Check if ignore_eos resulted in more tokens or exactly max_tokens
LOW	…nd/basic_function/offloading/test_npu_offload_modes.py	70	# Check if server is running (basic functionality test)
LOW	…nai_server/validation/test_openai_server_ignore_eos.py	75	# Check if ignore_eos resulted in more tokens or exactly max_tokens
LOW	test/registered/observability/test_tracing.py	77	# Check if it's a RequestStageConfig (has stage_name and level attributes)
LOW	…stered/amd/disaggregation/test_disaggregation_basic.py	322	# Check if servers are still healthy
LOW	…t/registered/amd/accuracy/mi30x/test_gsm8k_eval_amd.py	286	# Print results
LOW	…gistered/amd/accuracy/mi30x/test_vlms_mmmu_eval_amd.py	255	# Print results
LOW	…amd/accuracy/mi35x/test_qwen3_coder_next_eval_mi35x.py	218	# Check if model exists
LOW	…racy/mi35x/test_deepseek_r1_mxfp4_kv_fp8_eval_mi35x.py	197	# Check if model exists
LOW	…md/accuracy/mi35x/test_deepseek_r1_mxfp4_eval_mi35x.py	194	# Check if model exists
LOW	…y/mi35x/test_deepseek_r1_mxfp4_ar_fusion_eval_mi35x.py	196	# Check if model exists
LOW	…/registered/rl/test_update_weights_from_distributed.py	642	# Check if the weights of lm_head are tied with embed_tokens.
LOW	python/sglang/bench_serving.py	1309	# Check if at least one warmup request succeeded
LOW	python/sglang/utils.py	414	# Check if the cache file already exists
LOW	python/sglang/bench_one_batch.py	662	# Print output texts
LOW	python/sglang/benchmark/utils.py	106	# Check if the cache file already exists
LOW	python/sglang/test/few_shot_gsm8k_engine.py	135	# Print results
LOW	python/sglang/test/bench_one_batch_server_internal.py	732	# Print results
LOW	python/sglang/test/send_one.py	230	# Print results
LOW	python/sglang/test/runners.py	497	# Check if the text is empty or only whitespace.
LOW	python/sglang/test/runners.py	792	# Check if the text is empty or only whitespace.
LOW	python/sglang/test/few_shot_gsm8k.py	138	# Print results
LOW	python/sglang/test/ci/ci_utils.py	238	# Check if we should retry
LOW	…/sglang/test/server_fixtures/disaggregation_fixture.py	228	# Check if port is active
LOW	…sglang/jit_kernel/benchmark/bench_custom_all_reduce.py	374	# Print results on rank 0.
LOW	python/sglang/jit_kernel/tests/test_pos_enc.py	480	# Print results
LOW	python/sglang/srt/dllm/mixin/scheduler.py	245	# Check if batch is full
LOW	python/sglang/srt/disaggregation/decode.py	1548	# Check if request was aborted due to corruption
LOW	python/sglang/srt/layers/sampler.py	392	# Check if any requests actually need logprobs computation
LOW	python/sglang/srt/layers/attention/xpu_backend.py	462	# Check if we should use local attention
LOW	python/sglang/srt/layers/attention/triton_backend.py	185	# Set static_kv_splits to False to use deterministic logic instead
LOW	…thon/sglang/srt/layers/attention/trtllm_mla_backend.py	749	# Check if we're in CUDA graph mode (buffers are pre-allocated)
LOW	python/sglang/srt/layers/attention/dsa_backend.py	579	# Check if MHA FP8 dequantization is needed
LOW	python/sglang/srt/layers/attention/dsa_backend.py	2249	# Check if sequence meets criteria for MHA_ONE_SHOT
LOW	…/sglang/srt/layers/attention/flashattention_backend.py	727	# Check if we should use local attention
LOW	…/sglang/srt/layers/attention/flashattention_backend.py	2981	# Check if we should use the specialized fast path for page_size=1, no SWA
LOW	…thon/sglang/srt/layers/attention/flashinfer_backend.py	1140	# Check if this specific wrapper's begin_forward has been replaced with fast_decode_plan
LOW	…ng/srt/layers/attention/triton_ops/extend_attention.py	864	# Check if we can skip this tile
206 more matches not shown…

Verbosity Indicators239 hits · 390 pts

Severity	File	Line	Snippet
LOW	benchmark/bench_linear_attention/bench_gdn_decode.py	104	# Step 1: split (same as forward_decode)
LOW	benchmark/bench_linear_attention/bench_gdn_decode.py	107	# Step 2: view + reshape (same as forward_decode)
LOW	benchmark/bench_linear_attention/bench_gdn_decode.py	112	# Step 3: fused gating + recurrent update
LOW	…mark/bench_linear_attention/bench_fused_gate_cumsum.py	79	# Step 1: gate activation using torch ops
LOW	…mark/bench_linear_attention/bench_fused_gate_cumsum.py	87	# Step 2: chunk-local cumsum
LOW	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	38	# Step 1: Create kind cluster (idempotent)
LOW	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	50	# Step 2: Build Docker images (unless SKIP_DOCKER_BUILD=1)
LOW	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	75	# Step 3: Load images into kind
LOW	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	82	# Step 4: Apply namespace and RBAC
LOW	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	89	# Step 5: Deploy 3 fake-worker replicas behind a Service
LOW	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	144	# Step 6: Create sgl-router ConfigMap with k8s discovery pointing at the
LOW	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	175	# Step 7: Deploy sgl-router
LOW	test/srt/cpu/utils.py	72	# Step 2: compute per-block max abs values → scale
LOW	test/manual/test_tokenizer_manager.py	336	# Step 1: Detect format
LOW	test/manual/test_tokenizer_manager.py	342	# Step 2: Prepare input
LOW	test/manual/test_tokenizer_manager.py	348	# Step 3: Extract results (simulated tokenizer output)
LOW	test/manual/test_tokenizer_manager.py	367	# Step 1: Detect format
LOW	test/manual/test_tokenizer_manager.py	373	# Step 2: Prepare input
LOW	test/manual/test_tokenizer_manager.py	379	# Step 3: Extract results (simulated tokenizer output for cross-encoder)
LOW	test/manual/test_tokenizer_manager.py	396	# Step 1: Detect format
LOW	test/manual/test_tokenizer_manager.py	402	# Step 2: Prepare input
LOW	test/manual/test_tokenizer_manager.py	408	# Step 3: Extract results (simulated tokenizer output)
LOW	test/manual/dsv4/test_fused_compress_attn_hip.py	312	# Step 1: write current tokens to state (same for both paths)
LOW	test/registered/unit/managers/test_hisparse_unit.py	446	# Step 1: load the first TOP_K positions from host (no newest token —
LOW	test/registered/unit/managers/test_hisparse_unit.py	455	# Step 2: half overlap (hit) + half new (miss).
LOW	test/registered/unit/parser/test_reasoning_parser.py	1147	# Step 1: Send partial end tag when not in reasoning mode
LOW	test/registered/unit/parser/test_reasoning_parser.py	1153	# Step 2: Send normal text that doesn't complete the end tag
LOW	…egistered/unit/mem_cache/test_decode_radix_lock_ref.py	128	# Step 1: inc_lock_ref (pop_preallocated locks the matched node)
LOW	…egistered/unit/mem_cache/test_decode_radix_lock_ref.py	144	# Step 2: cache_unfinished_req (dec old lock, inc new lock)
LOW	…egistered/unit/mem_cache/test_decode_radix_lock_ref.py	147	# Step 3: cache_finished_req with is_insert=True (dec lock)
LOW	…egistered/unit/mem_cache/test_decode_radix_lock_ref.py	176	# Step 1: inc_lock_ref on root (simulates get_new_prebuilt_batch)
LOW	…egistered/unit/mem_cache/test_decode_radix_lock_ref.py	193	# Step 2: cache_unfinished_req (dec root=no-op, inc new leaf)
LOW	…egistered/unit/mem_cache/test_decode_radix_lock_ref.py	196	# Step 3: cache_finished_req (dec leaf)
LOW	test/registered/unit/mem_cache/test_mamba_unittest.py	684	# Step 1: Insert [1,2,3] to create first node
LOW	test/registered/unit/mem_cache/test_mamba_unittest.py	696	# Step 2: Insert [1,2,3,4,5,6,7] with prev_prefix_len=0 (free all matched)
LOW	test/registered/unit/mem_cache/test_mamba_unittest.py	713	# Step 3: Insert [1,2,3,4,5,6,7,8] with prev_prefix_len=2
LOW	test/registered/unit/mem_cache/test_mamba_unittest.py	731	# Step 4: Insert [1,2,3,4,5,6,7,8,9] with prev_prefix_len=8 (covers all matched)
LOW	…ed/unit/mem_cache/test_unified_radix_cache_unittest.py	637	# Step 1: insert 1 page
LOW	…ed/unit/mem_cache/test_unified_radix_cache_unittest.py	641	# Step 2: insert 2 pages with prev_prefix_len=0 → frees overlap of 1 page
LOW	…ed/unit/mem_cache/test_unified_radix_cache_unittest.py	659	# Step 3: insert 3 pages with prev_prefix_len=len(seq_2p) → nothing freed
LOW	test/registered/cpu/utils.py	72	# Step 2: compute per-block max abs values → scale
LOW	test/registered/function_call/test_kimik2_detector.py	549	# Step 1: reasoning parser
LOW	test/registered/function_call/test_kimik2_detector.py	553	# Step 2: feed normal_text into tool call parser (like serving_chat.py does)
LOW	test/registered/function_call/test_kimik2_detector.py	585	# Step 1: reasoning parser
LOW	test/registered/function_call/test_kimik2_detector.py	590	# Step 2: tool call parser on normal_text
LOW	test/registered/debug_utils/test_dumper.py	4090	# Step 1: graft input. target sends its real q to baseline along
LOW	test/registered/debug_utils/test_dumper.py	4099	# Step 2: target runs the (suspected buggy) attention kernel —
LOW	test/registered/debug_utils/test_dumper.py	4103	# Step 3: graft output. baseline sends its (good) attn_out to
LOW	test/registered/debug_utils/test_dumper.py	4064	# Step 1: graft input. target sends its q to baseline; baseline's
LOW	test/registered/debug_utils/test_dumper.py	4075	# Step 2: baseline runs the known-good attention kernel.
LOW	test/registered/debug_utils/test_dumper.py	4078	# Step 3: graft output. baseline sends attn_out to target with a
LOW	…t/registered/debug_utils/comparator/test_entrypoint.py	2206	# Step 0: prefill with 2 sequences (3+2 tokens)
LOW	…t/registered/debug_utils/comparator/test_entrypoint.py	2215	# Step 1: decode (1 token per sequence)
LOW	…t/registered/debug_utils/comparator/test_entrypoint.py	2267	# Step 0: prefill — seq A (3 tokens) + seq B (2 tokens)
LOW	…t/registered/debug_utils/comparator/test_entrypoint.py	4720	# Step 1: Create 4 target ranks where moe_tp is absent from ranks 2-3.
LOW	…t/registered/debug_utils/comparator/test_entrypoint.py	4723	# Step 2: _is_jointly_determined is called with parent_axes={tp, moe_tp}
LOW	…t/registered/debug_utils/comparator/test_entrypoint.py	4725	# Step 3: edp remains undeclared → ValueError emitted as error record.
LOW	…t/registered/debug_utils/comparator/test_entrypoint.py	4822	# Step 1: 4 target ranks with TP=2, CP=2 (replicated), EDP=4.
LOW	…t/registered/debug_utils/comparator/test_entrypoint.py	4824	# Step 2: _is_jointly_determined is called with parent_axes={tp, cp}, child=edp.
LOW	…t/registered/debug_utils/comparator/test_entrypoint.py	4826	# Step 3: CP replicated picks one rank per tp group → TP concat → correct shape.
179 more matches not shown…

Docstring Block Structure32 hits · 160 pts

Severity	File	Line	Snippet
HIGH	python/sglang/test/ascend/test_ascend_utils.py	467	Start the service and obtain the inference results. Parameters: model: Model name num_prompts: Tota
HIGH	python/sglang/srt/server_args_config_parser.py	52	Merge configuration file arguments with command-line arguments. Configuration arguments are inserted a
HIGH	python/sglang/srt/server_args_config_parser.py	101	Parse YAML configuration file and convert to argument list. Args: file_path: Path to the Y
HIGH	python/sglang/srt/layers/model_parallel.py	29	Locally shards a full tensor based on indicated sharding arrangement, and returns a DTensor containing the loca
HIGH	…thon/sglang/srt/layers/attention/flashinfer_backend.py	335	Process multi-item scoring tensors for FlashInfer attention. This method handles sequences containing multiple
HIGH	python/sglang/srt/layers/attention/fla/chunk.py	146	Args: q (torch.Tensor): queries of shape `[B, T, H, K]` if `head_first=False` else `[B, H, T, K
HIGH	…hon/sglang/srt/layers/attention/fla/fused_recurrent.py	459	Args: q (torch.Tensor): queries of shape `[B, T, H, K]`. k (torch.Tensor):
HIGH	python/sglang/srt/layers/quantization/fp8_kernel.py	1638	Quantize input tensor to FP8 (8-bit floating point) format. Args: input (torch.Tensor): Input tensor to be quantiz
HIGH	python/sglang/srt/layers/moe/cutlass_moe.py	55	Performs Fused MoE computation using CUTLASS-like kernels with FP8 weights and activations. This function implement
HIGH	python/sglang/srt/function_call/utils.py	223	Get consolidated $defs from all tools, validating for conflicts. Args: tools: List of tools to process
HIGH	python/sglang/srt/utils/network.py	119	Bind a TCP socket on the first available address family (IPv4/IPv6). Iterates over address families returned by _ge
HIGH	python/sglang/srt/utils/network.py	309	Automatically detect the local IP address using multiple fallback strategies. This function attempts to obtain
HIGH	python/sglang/srt/utils/nvtx_pytorch_hooks.py	46	Descends iterators that contains Tensors and prints the Tensor Recursive function that descends iterator type a
HIGH	python/sglang/srt/utils/nvtx_pytorch_hooks.py	74	Extract the static parameters from LLM and VLM relevant layer types Args: module_obj(class): Module
HIGH	python/sglang/srt/utils/nvtx_pytorch_hooks.py	198	Callback function that ends the NVTX marker Records the module name and tensor information Called after
HIGH	python/sglang/srt/utils/nvtx_pytorch_hooks.py	218	Creates an NVTX marker with the module name in it. This function is called before the module executes
HIGH	python/sglang/srt/utils/nvtx_pytorch_hooks.py	256	User level function that activates all the hooks The user needs to call this method from the network source cod
HIGH	python/sglang/srt/models/minicpmo.py	134	In streaming audio generation, determine which `text` positions the TTS model can attend to when generating each ch
HIGH	python/sglang/srt/models/minicpmo.py	613	Merge `input_ids` and `lm_spk_emb_last_hidden_states` to `inputs_embeds`. Args: input_ids (torch.Te
HIGH	python/sglang/srt/multimodal/processors/qwen_vl.py	133	calculate the number of frames for video used for model inputs. Args: ele (dict): a dict contains the confi
HIGH	python/sglang/srt/multimodal/processors/ernie45_vl.py	133	calculate the number of frames for video used for model inputs. Args: ele (dict): a dict contains the confi
HIGH	python/sglang/srt/mem_cache/storage/backend_factory.py	73	Create a storage backend instance. Args: backend_name: Name of the backend to create sto
HIGH	python/sglang/srt/model_loader/ci_weight_validation.py	1840	CI-specific download with validation and automatic retry on corruption. This function handles the download of
HIGH	python/sglang/srt/speculative/dflash_utils.py	164	Select target layer indices used to build DFlash context features. Args: num_target_layers: Number of trans
HIGH	python/sglang/srt/speculative/dflash_utils.py	427	Compute DFlash accept lengths and bonus tokens (greedy verify rule). Args: candidates: Token ids proposed b
HIGH	…/runtime/managers/memory_managers/layerwise_offload.py	424	Update consolidated CPU buffers with new weights. When layerwise offload (--dit-layerwise-offload) is enabled,
HIGH	…ang/multimodal_gen/runtime/utils/nvtx_pytorch_hooks.py	97	Walk ``model`` and attach forward pre/post hooks to every module. Args: model: Root module to instr
HIGH	…ng/multimodal_gen/runtime/models/encoders/qwen2_5vl.py	554	Calculate the 3D rope index based on image and video's temporal, height and width in LLM. Explanation:
HIGH	…imodal_gen/runtime/pipelines_core/stages/validators.py	345	Add a validation check for a field. Args: field_name: Name of the field being checked
HIGH	sgl-model-gateway/e2e_test/infra/run_eval.py	64	Run an evaluation and return metrics. Args: args: Configuration object with attributes: - base_
HIGH	sgl-model-gateway/e2e_test/infra/model_pool.py	727	Get a model instance by model_id, mode, and worker_type. If the model is not running, it will be launched on-de
HIGH	sgl-model-gateway/e2e_test/responses/test_basic_crud.py	24	Wait for background task to complete. Args: client: OpenAI client response_id: Response ID to poll

AI Slop Vocabulary40 hits · 98 pts

Severity	File	Line	Snippet
MEDIUM	docker/k8s-sglang-distributed-sts.yaml	58	privileged: true # to leverage RDMA/InfiniBand device, co-work with HostNetwork=true
MEDIUM	…t/manual/layers/attention/dsa/test_act_quant_triton.py	269	# Run comprehensive benchmark
MEDIUM	test/manual/layers/moe/bench_mxfp4_sm90_kernels.py	277	# Timing harness
MEDIUM	test/manual/4-gpu-models/test_qwen35_models_archived.py	17	# This eval harness applies the chat_template, which is critical for qwen3.5
MEDIUM	test/manual/4-gpu-models/test_qwen35_fp4_triton.py	5	# This eval harness applies the chat_template, which is critical for qwen3.5
MEDIUM	test/manual/eval/validate_longbench_v2_standalone.py	212	"""Generate comprehensive validation report."""
MEDIUM	…tered/attention/unittests/dense/test_flex_attention.py	65	# from logical positions, so it's robust to all non-tidy layouts.
MEDIUM	test/registered/attention/unittests/dense/test_fa3.py	51	# FlashAttention kernels are most stable in this harness with FA-friendly dims.
MEDIUM	test/registered/attention/unittests/dense/test_fa4.py	46	# FlashAttention kernels are most stable in this harness with FA-friendly dims.
MEDIUM	…egistered/attention/unittests/swa/test_torch_native.py	104	# arithmetic, so it's robust to all non-tidy layouts.
MEDIUM	test/registered/hicache/test_qwen35_hicache.py	15	# This eval harness applies the chat_template, which is critical for qwen3.5
LOW	test/registered/debug_utils/test_dumper.py	3052	# code can simply call `from sglang.srt.debug_utils.dumper import dumper`
MEDIUM	…attention_unittest/attention_methods/dsv4_attention.py	1255	# In runner-harness flows the reference is called BEFORE
LOW	python/sglang/srt/disaggregation/nixl/conn.py	101	), # hacky just add it into the message that will be sent
MEDIUM	python/sglang/srt/disaggregation/common/conn.py	929	# Enable higher PP ranks to be bootstrapped earlier to make PP PD requests bootstrap more robust
LOW	…/sglang/srt/layers/attention/flashattention_backend.py	3196	# then we can simply use a cdiv for the rest.
LOW	…ang/srt/layers/attention/mamba/causal_conv1d_triton.py	152	# first chunk and does not have prior-token, so just set to 0
MEDIUM	…hon/sglang/srt/layers/attention/dsa/tilelang_kernel.py	1102	# to better utilize FP8 dynamic range, then apply the inverse scale after GEMM.
MEDIUM	python/sglang/srt/layers/quantization/fpgemm_fp8.py	47	# For GPUs that lack FP8 hardware suspport, we can leverage the Marlin
MEDIUM	python/sglang/srt/layers/quantization/fp8.py	230	# Keep both "model." and non-"model." variants for robust prefix matching.
MEDIUM	python/sglang/srt/layers/quantization/fp8.py	336	# For GPUs that lack FP8 hardware support, we can leverage the Marlin
MEDIUM	python/sglang/srt/layers/quantization/fp8_utils.py	622	# TODO: add more robust shape check here
MEDIUM	…hon/sglang/srt/layers/quantization/marlin_utils_fp8.py	69	# For GPUs that lack FP8 hardware support, we can leverage the
LOW	…srt/layers/quantization/mxfp4_flashinfer_trtllm_moe.py	451	# in-place. Otherwise `routed` is already scale-final and we just add
LOW	python/sglang/srt/managers/io_struct.py	760	lora_id: Optional[str] = None # None means just use the base model
LOW	python/sglang/srt/managers/io_struct.py	1057	lora_id: Optional[str] = None # None means just use the base model
LOW	python/sglang/srt/managers/scheduler_pp_mixin.py	438	# otherwise, just pass along previous consensus
LOW	python/sglang/srt/managers/scheduler.py	3661	# In-place pause: just set the flag and return immediately.
LOW	python/sglang/srt/function_call/qwen3_coder_detector.py	41	# Base class already initializes _buffer, we just use it directly
LOW	python/sglang/srt/parser/harmony_parser.py	492	# If no emit, just return the held content
LOW	python/sglang/srt/mem_cache/hiradix_cache.py	1143	# unknown prefetch stop policy, just return True
MEDIUM	…n/sglang/multimodal_gen/test/server/accuracy_config.py	57	# of real divergence or unsupported reference loading in the harness.
LOW	python/sglang/multimodal_gen/test/server/gpu_cases.py	60	# To test different models, simply add more DiffusionCase entries
MEDIUM	…on/sglang/multimodal_gen/test/server/accuracy_hooks.py	26	# These are harness defaults for synthetic accuracy inputs.
MEDIUM	…lang/multimodal_gen/runtime/layers/quantization/fp8.py	191	# For GPUs that lack FP8 hardware support, we can leverage the Marlin
MEDIUM	…imodal_gen/runtime/pipelines_core/stages/validators.py	503	"""Get a comprehensive summary of all validation failures."""
MEDIUM	sgl-kernel/csrc/cpu/decode.cpp	1429	// use smaller BLOCK_H when batches is small to utilize all cores
MEDIUM	…del-gateway/e2e_test/k8s_integration/test_lifecycle.py	374	# "total count == 0" keeps the test robust if a parallel/earlier
MEDIUM	…way/e2e_test/chat_completions/test_function_calling.py	455	# Make the test more robust by checking type and accepting valid responses
MEDIUM	…l-gateway/bindings/python/src/sglang_router/mini_lb.py	440	# We may utilize `GenerateReqInput`'s logic later

Cross-Language Confusion14 hits · 60 pts

Severity	File	Line	Snippet
HIGH	benchmark/hicache/bench_mix.py	465	self.user_generator.push(
HIGH	experimental/sgl-router/tests/e2e/infra/gateway.py	353	"Build it first: `cd experimental/sgl-router && cargo build --release` "
HIGH	test/manual/models/test_qwen3_asr.py	630	f"first commit's previous_item_id must be JSON null, got {committed_1!r}",
HIGH	test/manual/debug_utils/test_log_parser.py	16	"""[{"line":"(SGLangEngine pid=35555) [2025-10-31 03:45:20 TP0] Decode batch [51341], #running-req: 317, #to
HIGH	…tion_unittest/runner_modes/cuda_graph_decode_runner.py	417	# capture, which makes the capture-time output undefined; only
HIGH	…t/layers/quantization/nvfp4_gemm_swiglu_nvfp4_quant.py	2552	:param c_sf_ptr: Pointer to scale factor tensor for C (can be null)
HIGH	…t/layers/quantization/nvfp4_gemm_swiglu_nvfp4_quant.py	2556	:param norm_const_ptr: Pointer to normalization constant for SFC generation (can be null)
HIGH	python/sglang/srt/function_call/utils.py	297	return "string" # If only null, default to string
HIGH	python/sglang/srt/function_call/mimo_detector.py	144	<parameter=command>pwd && ls</parameter>
HIGH	python/sglang/srt/utils/field_validators.py	66	raise ValueError(f"must be list or null; got {type(v).__name__}")
HIGH	python/sglang/srt/utils/hf_transformers/tokenizer.py	366	# When the config omits the key or has null, use the v4 default so that
HIGH	python/sglang/srt/entrypoints/http_server.py	429	{"error": {"message": "...", "type": "...", "param": null, "code": <status>}}
HIGH	…al-refactor-verify/mechanical_refactor_verify_utils.py	30	exec_command(f"git add -A && git commit -m {shlex.quote(message)}", cwd=cwd)
HIGH	scripts/ci/utils/runner_utilization_report.py	188	equal to `created_at` (not null). The previous code required both a

Hallucination Indicators6 hits · 60 pts

Severity	File	Line	Snippet
CRITICAL	…tention_unittest/attention_methods/mamba2_attention.py	537	self.mixer.norm.weight.data = self.mixer.norm.weight.data.float()
CRITICAL	python/sglang/srt/layers/attention/fla/utils.py	231	return triton.runtime.driver.active.utils.get_device_properties(tensor_idx)[
CRITICAL	python/sglang/srt/layers/attention/fla/utils.py	286	triton.runtime.driver.active.utils.get_device_properties(i)[
CRITICAL	python/sglang/srt/models/glm4_moe.py	490	) and self.shared_experts.gate_up_proj.quant_method.quant_config.get_name() in {
CRITICAL	python/sglang/srt/models/deepseek_v2.py	734	) and self.shared_experts.gate_up_proj.quant_method.quant_config.get_name() in {
CRITICAL	…l_gen/runtime/pipelines_core/composed_pipeline_base.py	124	task_name = self.server_args.pipeline_config.task_type.name.lower()

Slop Phrases18 hits · 42 pts

Severity	File	Line	Snippet
MEDIUM	…manual/openai_server/function_call/test_tool_choice.py	845	# # Start the local OpenAI Server. If necessary, you can add other parameters such as --enable-tools.
MEDIUM	…d/ascend/interface/test_npu_openai_function_calling.py	53	# Start the local OpenAI Server. If necessary, you can add other parameters such as --enable-tools.
MEDIUM	…i_server/function_call/test_openai_function_calling.py	43	# Start the local OpenAI Server. If necessary, you can add other parameters such as --enable-tools.
MEDIUM	…i_server/function_call/test_openai_function_calling.py	926	# # Start the local OpenAI Server. If necessary, you can add other parameters such as --enable-tools.
LOW	python/sglang/test/runners.py	488	# make sure to disable compile
MEDIUM	python/sglang/jit_kernel/flash_attention_v3.py	92	# That means if you use A100/A*0/L20/L40/L40s/4090 you can use fa3.
LOW	python/sglang/srt/models/hunyuan.py	784	# make sure to leave KV cache scale factors in a known good (dummy) state
LOW	python/sglang/srt/models/apertus.py	384	# make sure to leave KV cache scale factors in a known good (dummy) state
LOW	python/sglang/srt/models/mimo_v2.py	968	# make sure to leave KV cache scale factors in a known good (dummy) state
LOW	python/sglang/srt/models/llama.py	436	# make sure to leave KV cache scale factors in a known good (dummy) state
LOW	python/sglang/srt/models/glm4.py	398	# make sure to leave KV cache scale factors in a known good (dummy) state
LOW	python/sglang/srt/models/qwen2.py	403	# make sure to leave KV cache scale factors in a known good (dummy) state
LOW	…glang/multimodal_gen/runtime/models/vaes/hunyuanvae.py	191	# If you are encountering an error here, make sure to try running encoding/decoding with
MEDIUM	sgl-kernel/python/sgl_kernel/flash_attn.py	24	# That means if you use A100/A*0/L20/L40/L40s/4090 you can use fa3.
MEDIUM	sgl-kernel/tests/test_flash_attention.py	28	# That means if you use A100/A*0/L20/L40/L40s/4090 you can use fa3.
MEDIUM	…ntend_language/quick_start/local_example_llava_next.py	55	# Or you can use the 72B model
MEDIUM	…ntend_language/quick_start/local_example_llava_next.py	62	# Or you can use API models
MEDIUM	…-gateway/examples/wasm/wasm-guest-ratelimit/src/lib.rs	37	// This is a simplified example for demonstration purposes

Dead Code20 hits · 40 pts

Severity	File	Line
MEDIUM	python/sglang/srt/entrypoints/openai/serving_base.py	146
MEDIUM	python/sglang/srt/entrypoints/openai/serving_base.py	149
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	706
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	707
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	708
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	711
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	712
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	714
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	716
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	719
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	720
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	722
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	732
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	747
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	757
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	758
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	759
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	787
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	788
MEDIUM	…ang/multimodal_gen/runtime/models/dits/hunyuanvideo.py	790

Synthetic Comment Markers2 hits · 15 pts

Severity	File	Line	Snippet
HIGH	test/manual/attention/test_trtllm_mla_backend.py	1238	# Reshape as requested
HIGH	…ang/multimodal_gen/runtime/utils/hf_diffusers_utils.py	413	# Special handling for stop token <\|eom_id\|> generated by llama 3 tool use.

Fake / Example Data12 hits · 12 pts

Severity	File	Line	Snippet
LOW	…l/sgl-router/tests/scripts/generate_parity_fixtures.py	41	"Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod "
LOW	…l/sgl-router/tests/scripts/generate_parity_fixtures.py	41	"Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod "
LOW	python/sglang/test/kits/ebnf_constrained_kit.py	62	self.__class__.ebnf_grammar = 'root ::= "user@example.com"'
LOW	…-model-gateway/tests/security/auth_integration_test.rs	174	let claims = create_claims("admin@example.com", vec!["admin"]);
LOW	…-model-gateway/tests/security/auth_integration_test.rs	187	assert_eq!(validated.subject, "admin@example.com");
LOW	…-model-gateway/tests/security/auth_integration_test.rs	213	let claims = create_claims("user@example.com", vec!["user", "viewer"]);
LOW	…-model-gateway/tests/security/auth_integration_test.rs	221	assert_eq!(validated.subject, "user@example.com");
LOW	…-model-gateway/tests/security/auth_integration_test.rs	252	sub: "user@example.com".to_string(),
LOW	…-model-gateway/tests/security/auth_integration_test.rs	301	sub: "user@example.com".to_string(),
LOW	…-model-gateway/tests/security/auth_integration_test.rs	338	let claims = create_claims("user@example.com", vec!["admin"]);
LOW	…-model-gateway/tests/security/auth_integration_test.rs	590	let claims = create_claims("user@example.com", vec!["admin"]);
LOW	…-model-gateway/tests/security/auth_integration_test.rs	713	let claims = create_claims("user@example.com", vec!["admin"]);

Example Usage Blocks5 hits · 6 pts

Severity	File	Line	Snippet
LOW	…rimental/sgl-router/tests/e2e/k8s_integration/setup.sh	6	# Usage:
LOW	scripts/killall_sglang.sh	8	# Usage:
LOW	scripts/ci/musa/rename_wheels_musa.sh	10	# Usage:
LOW	sgl-model-gateway/e2e_test/k8s_integration/setup.sh	8	# Usage:
LOW	…ngs/golang/examples/oai_server/scripts/analyze_tpot.sh	6	# Usage:

Overly Generic Function Names4 hits · 4 pts

Severity	File	Line	Snippet
LOW	benchmark/hicache/bench_mix.py	417	async def handle_request(self, user_data):
LOW	benchmark/hicache/bench_multiturn.py	388	async def handle_request(self, item):
LOW	python/sglang/srt/debug_utils/dumper.py	1329	def handle_request(self, *, method: str, body: dict[str, Any]) -> list[dict]:
LOW	python/sglang/srt/entrypoints/openai/serving_base.py	73	async def handle_request(

Score History

Severity Breakdown

Pattern Findings