flashinfer-ai/flashinfer

20.0

Adjusted Score

20.0

Raw Score

100%

Time Factor

2026-07-14

Last Push

6.0K

Stars

Python

Language

667.3K

Lines of Code

1.4K

Files

7.7K

Pattern Hits

2026-07-14

Scan Date

0.31

HC Hit Rate

What These Metrics Mean

Adjusted Score: Primary synthetic code indicator. Raw score normalised per 1,000 lines of code and multiplied by the temporal discount factor. This is the definitive comparative metric — use it to rank repositories by AI authorship density.
Raw Score: The unmodified sum of all severity-weighted, context-multiplied pattern match scores before temporal discounting. Reflects the absolute signal strength independent of when the repository was last active.
Time Factor: The temporal discount multiplier (0–100%) applied to the raw score. Repositories last updated before ChatGPT's launch (Nov 2022) receive a 5% factor. Full signal is only assigned to repositories active in the post-adoption era (Jan 2024+).
Pattern Hits: Total count of individual pattern matches across all files and categories. A high hit count with a low score may indicate a very large codebase with isolated AI snippets; a low count with a high score indicates dense, concentrated AI signatures.
HC Hit Rate: High+Critical pattern hits per file, averaged across the repository. This orthogonal signal catches repositories where a few files are densely packed with high-severity AI tells — a strong indicator even when the normalised score appears moderate due to codebase size.
Lines of Code / Files: Total lines and files analysed. The scanner examines 94 file extensions. These denominators are used to normalise the score, enabling fair comparison between repositories of vastly different sizes.

Score History

This chart maps the temporal evolution of the adjusted synthetic code score across successive scan runs. An upward trajectory indicates ongoing incorporation of AI-generated code or expanding LLM-assisted scaffolding; a stable or declining trajectory may reflect active human refactoring, code removal, or the adoption of stricter authorship policies. The dashed secondary line (right axis) independently tracks total raw pattern hit count, which can diverge from the normalised score when codebase size changes significantly between scans.

Severity Breakdown

Classifies detected patterns by their diagnostic confidence and structural impact. CRITICAL patterns (coefficient 10) represent definitive synthetic signatures — hallucinated imports, explicit LLM attribution metadata — virtually never produced by human authors. HIGH (5) indicates strong structural tells such as cross-file repetition or cross-linguistic idioms. MEDIUM (2) covers recognisable conversational padding and AI-specific vocabulary. LOW (1) captures subtle indicators like tautological comments and generic boilerplate that require density to carry independent signal.

CRITICAL 0HIGH 450MEDIUM 2160LOW 5125

Directory Score Breakdown

This horizontal bar chart decomposes the repository's raw synthetic code score by top-level directory, allowing you to pinpoint precisely which modules or components carry the highest AI authorship density. Directories with disproportionately high scores relative to their size warrant targeted manual review: concentrated AI signatures often trace back to mass-generated configuration layers, auto-ported test suites, LLM-scaffolded boilerplate classes, or entire subsystems authored under heavy copilot assistance. Use this view to prioritise your human code-review effort.

Pattern Findings

The scanner identified 7735 distinct pattern matches across 22 syntactic categories. Each entry below represents a discrete location in the source code where the engine recorded a statistically significant AI authorship indicator. Expand any category row to inspect the individual file paths, line numbers, code snippets, and the lexical context (CODE, COMMENT, or STRING) in which each match was detected.

Reading the findings table: The Severity column indicates the diagnostic confidence level (CRITICAL / HIGH / MEDIUM / LOW). The Context column identifies whether the match occurred inside executable code, an inline comment, or a string literal — comment-context matches receive a ×1.5 weight because LLMs systematically over-annotate. The ⚡ bolt icon marks clustered matches: three or more patterns within a 10-line window, each receiving an additional ×1.5 density multiplier as dense clusters constitute far stronger evidence of synthetic authorship than isolated hits.

Decorative Section Separators2049 hits · 6183 pts

Severity	File	Line	Snippet	Context
MEDIUM	csrc/fused_moe/monomoe/src/ptx_utils.h	14	// ── Hopper WGMMA (sm_90a) helpers ─────────────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/ptx_utils.h	175	// ── Hopper mbarrier helpers (sm_90a) ──────────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/ptx_utils.h	321	// ── Hopper TMA bulk-tensor copy (sm_90a) ──────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_tma.h	214	// ─── Device-side TMA load helpers ────────────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_tma.h	437	// ─── Down-projection (Phase 4) TMA load helpers ──────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	19	// ── Default pipeline variants ───────────────────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	66	// ── KernelConfig accessors ───────────────────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	106	// ── WGMMA-path-only up-proj → down-proj scratchpad ───────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	182	// ── Block-wise quantization detection (forward declaration) ──────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	224	// ── WGMMA opt-in detection ───────────────────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	235	// ── TMA opt-in detection ────────────────────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	262	// ── WGMMA-only tile dimensions (v1 dual-WG K=128 streaming) ──────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	354	// ── WGMMA down-projection grid layout ────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	491	// ── K-substep constants ────────────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	548	// ── NEW: single-buffer fp8_act covering all K substeps ──────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	744	// ── TMA-only extensions ─────────────────────────────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	789	// ── routing-window mbarrier ─────────────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	806	// ── Phase 3 → Phase 4 (expert, token) reorganization tables ─────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	884	// ── Routing-time inverse map: (expert id, token) → top-k rank ────────	COMMENT
MEDIUM	csrc/fused_moe/monomoe/src/moe_internal.h	941	// ── Aliasing safety static_asserts ─────────────────	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	745	# ==============================================================================	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	747	# ==============================================================================	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1263	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1265	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1384	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1386	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1460	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1462	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1492	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1494	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1895	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1897	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1941	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	1943	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2011	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2013	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2071	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2073	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2173	# ==============================================================================	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2175	# ==============================================================================	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2184	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2186	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2256	# ------------------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/kda/test_recurrent_kda.py	2258	# ------------------------------------------------------------------------------	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	36	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	38	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	106	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	108	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	127	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	129	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	506	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	508	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	900	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	902	# ==============================================================================	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	1062	# ------------------------------------------------------------------------------	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	1064	# ------------------------------------------------------------------------------	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	1165	# ------------------------------------------------------------------------------	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	1167	# ------------------------------------------------------------------------------	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	2107	# ------------------------------------------------------------------------------	COMMENT
MEDIUM	tests/kda/test_recurrent_kda.py	2109	# ------------------------------------------------------------------------------	COMMENT
1989 more matches not shown…

Cross-File Repetition410 hits · 2050 pts

Severity	File	Line	Snippet	Context
HIGH	build_backend.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_batch_prefill_kernels.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_alibi.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_batch_decode_kernels.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_shared_prefix_kernels.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_decode_fp8_calibration_scale.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_deepseek_mla.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/utils/test_pod_kernels.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/test_helpers/jit_utils.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	benchmarks/routines/__init__.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/version.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/pod.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/__init__.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/page.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/utils.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/prefill.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/decode.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/cascade.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/comm/nvshmem_allreduce.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/testing/__init__.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer/testing/utils.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	build_utils.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer-cubin/flashinfer_cubin/__init__.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_attention_sink_blackwell.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_attention_sink.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_batch_attention.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/attention/test_vsa_block_sparse.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/trace/test_fi_trace.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/mamba/__init__.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/utils/test_logging.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/utils/test_fp4_kv_quantization.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/utils/test_logging_replay.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/utils/test_load_cubin_compile_race_condition.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/gemm/test_groupwise_scaled_gemm_fp8.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/gemm/test_groupwise_scaled_gemm_mxfp4.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/gdn/conftest.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/gdn/test_prefill_cp_delta_rule.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/gdn/test_prefill_delta_rule.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/gdn/test_decode_delta_rule.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/gdn/reference_delta_rule.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	…sts/gdn/test_decode_pretranspose_noncontiguous_pool.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/test_helpers/sink_attention_reference.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/moe/test_trtllm_cutlass_fused_moe.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	…e/test_trtllm_gen_fused_moe_routing_renormalize_fp8.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/moe/trtllm_gen_fused_moe_utils.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	…/test_trtllm_gen_fused_moe_routing_renormalize_bf16.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/moe/test_cute_dsl_fused_moe.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/moe/utils.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	…e/test_trtllm_gen_fused_moe_routing_renormalize_fp4.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/moe/test_trtllm_gen_routed_fused_moe.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/moe/test_trtllm_gen_per_token_moe.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/moe/test_bgmv_moe.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/moe/test_trtllm_gen_fused_moe.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	tests/grouped_mm/test_cute_sm120_mxfp8.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer-jit-cache/build_backend.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	flashinfer-jit-cache/flashinfer_jit_cache/__init__.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	benchmarks/bench_deepgemm_blackwell.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	benchmarks/bench_gdn_decode.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	benchmarks/bench_gdn_prefill.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
HIGH	benchmarks/bench_groupwise_gemm_fp8_blackwell.py	0	copyright (c) 2025 by flashinfer team. licensed under the apache license, version 2.0 (the "license"); you may not use t	STRING
350 more matches not shown…

Hyper-Verbose Identifiers2039 hits · 1824 pts

Severity	File	Line	Snippet	Context
LOW	build_backend.py	317	def _synthesize_nccl_builddir(build: Path) -> None:	CODE
LOW	build_backend.py	375	def _check_nccl_version_drift(wheel: Path) -> None:	CODE
LOW	build_backend.py	623	def _install_nvep_runtime_wheels(built_nixl: bool) -> None:	CODE
LOW	build_backend.py	690	def _install_cuda_tile_compile_deps() -> None:	CODE
LOW⚡	build_backend.py	1027	def get_requires_for_build_wheel(config_settings=None):	CODE
LOW⚡	build_backend.py	1032	def get_requires_for_build_sdist(config_settings=None):	CODE
LOW⚡	build_backend.py	1037	def get_requires_for_build_editable(config_settings=None):	CODE
LOW⚡	build_backend.py	1042	def prepare_metadata_for_build_wheel(metadata_directory, config_settings=None):	CODE
LOW⚡	build_backend.py	1047	def prepare_metadata_for_build_editable(metadata_directory, config_settings=None):	CODE
LOW	tests/conftest.py	13	def _patch_cutlass_dsl_operand_major_mode():	CODE
LOW	tests/conftest.py	81	def _set_torch_compile_options():	CODE
LOW	tests/conftest.py	87	def _monkeypatch_add_torch_compile(func):	CODE
LOW	tests/conftest.py	175	def pytest_collection_modifyitems(config, items):	CODE
LOW	tests/test_artifacts.py	12	def test_sanity_check_urllib_behavior():	CODE
LOW	tests/test_artifacts.py	216	def _mock_file_index_responses():	STRING
LOW	tests/test_artifacts.py	230	def test_get_available_cubin_files():	STRING
LOW	tests/test_artifacts.py	282	def test_get_subdir_file_list(monkeypatch, tmp_path):	CODE
LOW⚡	tests/test_jit_cpp_ext.py	18	def test_nvcc_parallelism_flags_use_flashinfer_nvcc_threads(monkeypatch):	CODE
LOW⚡	tests/test_jit_cpp_ext.py	24	def test_nvcc_parallelism_flags_ignore_sccache_launcher(monkeypatch):	CODE
LOW⚡	tests/test_jit_cpp_ext.py	31	def test_generate_ninja_uses_sccache_compatible_nvcc_depfile_flag(	CODE
LOW	tests/test_jit_cpp_ext.py	51	def test_debug_jit_uses_sccache_compatible_nvcc_device_debug_flag(monkeypatch):	CODE
LOW	tests/test_jit_cpp_ext.py	69	def test_release_jit_propagates_ndebug_to_host_cflags(monkeypatch):	CODE
LOW	tests/test_jit_cpp_ext.py	88	def test_debug_jit_does_not_propagate_ndebug(monkeypatch):	CODE
LOW	tests/test_jit_cpp_ext.py	106	def test_run_ninja_uses_max_jobs(monkeypatch, tmp_path):	CODE
LOW	tests/test_jit_cpp_ext.py	132	def test_jit_spec_build_rewrites_ninja_before_build(monkeypatch):	CODE
LOW	tests/test_jit_cpp_ext.py	232	def fake_single_prefill_module(	CODE
LOW	tests/test_jit_cpp_ext.py	244	def fake_batch_prefill_module(	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	757	def test_non_contiguous_gate_stride(H: int, D: int):	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	1269	def test_spec_decode_padded_cuda_graph(D):	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	1389	def test_spec_decode_shape_validation():	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	1465	def test_spec_decode_cu_seqlens_validation():	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	1497	def test_spec_decode_num_accepted_tokens():	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	1900	def test_spec_decode_checkpoint_correctness():	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	1946	def test_spec_decode_non_compact_state():	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	2016	def test_spec_decode_all_padded():	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	2076	def test_spec_decode_validate_slots_collision():	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	2189	def test_spec_decode_batched_vs_cuseqlens():	CODE
LOW⚡	tests/kda/test_recurrent_kda.py	2261	def test_spec_decode_batched_auto_ssi():	CODE
LOW	tests/kda/test_recurrent_kda.py	91	def naive_kda_lowerbound_gate(	CODE
LOW	tests/kda/test_recurrent_kda.py	146	def _make_vs_reference_tensors(B, T, H, D, gate_logit_normalizer, dtype):	CODE
LOW	tests/kda/test_recurrent_kda.py	176	def test_recurrent_kda_vs_naive(	CODE
LOW	tests/kda/test_recurrent_kda.py	235	def test_recurrent_kda_vs_fla(	CODE
LOW	tests/kda/test_recurrent_kda.py	442	def test_standard_decode_state_indices_update_pool():	CODE
LOW	tests/kda/test_recurrent_kda.py	521	def test_vllm_padded_cuda_graph(	CODE
LOW	tests/kda/test_recurrent_kda.py	669	def test_non_compact_state_stride(H: int, D: int):	CODE
LOW	tests/kda/test_recurrent_kda.py	944	def spec_decode_naive_reference(	CODE
LOW	tests/kda/test_recurrent_kda.py	1178	def test_spec_decode_gate_modes(gate_mode):	CODE
LOW	tests/kda/test_recurrent_kda.py	1589	def test_spec_decode_nat_equals_one_matches_no_nat():	CODE
LOW	tests/kda/test_recurrent_kda.py	1633	def test_spec_decode_nat_heterogeneous(D):	CODE
LOW	tests/kda/test_recurrent_kda.py	1695	def test_spec_decode_nat_with_padding():	CODE
LOW	tests/kda/test_recurrent_kda.py	1815	def test_spec_decode_nat_gate_modes(gate_mode):	CODE
LOW	tests/attention/test_nvfp4_attention_sm120.py	24	def _patch_cutlass_dsl_operand_major_mode():	CODE
LOW	tests/attention/test_hopper.py	66	def test_batch_ragged_prefill(	CODE
LOW	tests/attention/test_trtllm_gen_mla.py	22	def trtllm_gen_workspace_softmax_end_bytes_decode(	CODE
LOW	tests/attention/test_trtllm_gen_mla.py	111	def sparse_mla_reference_torch(	CODE
LOW	tests/attention/test_trtllm_gen_mla.py	151	def scaled_dot_product_attention(	CODE
LOW	tests/attention/test_trtllm_gen_mla.py	662	def trtllm_batch_decode_mla_sparse(	CODE
LOW	tests/attention/test_trtllm_gen_mla.py	962	def test_trtllm_batch_decode_mla(	CODE
LOW	tests/attention/test_trtllm_gen_mla.py	1009	def test_trtllm_batch_decode_mla_cum_seq_lens_q(	CODE
LOW	tests/attention/test_trtllm_gen_mla.py	1044	def test_trtllm_batch_decode_mla_sparse(	CODE
1979 more matches not shown…

Unused Imports1214 hits · 1008 pts

Severity	File	Line	Context
LOW	tests/attention/test_sparse_mla_sm120.py	31	CODE
LOW	tests/attention/test_rope.py	19	CODE
LOW	tests/trace/reference_utils.py	3	CODE
LOW	tests/trace/test_fi_trace_template_consistency.py	411	CODE
LOW	tests/moe_ep/smoke_nixl_ep.py	12	CODE
LOW	tests/moe_ep/test_fleet_create.py	3	CODE
LOW	tests/moe_ep/conftest.py	3	CODE
LOW	tests/moe_ep/test_compute_bridge.py	18	CODE
LOW	tests/moe_ep/test_layer_factory.py	3	CODE
LOW	tests/moe_ep/test_moe_ep_compute_correctness.py	11	CODE
LOW	tests/moe_ep/test_moe_ep_compute_correctness_nvfp4.py	12	CODE
LOW	tests/moe_ep/test_split_kernels.py	7	CODE
LOW	tests/moe_ep/test_constraints.py	3	CODE
LOW	tests/moe_ep/test_moe_ep_ht_correctness.py	11	CODE
LOW	tests/moe_ep/test_moe_ep_deep_gemm_mega_multirank.py	14	CODE
LOW	tests/moe_ep/test_config.py	3	CODE
LOW	tests/moe_ep/test_fused_moe_weights.py	3	CODE
LOW	tests/moe_ep/test_moe_ep_layer_multirank.py	12	CODE
LOW	…sts/moe_ep/test_moe_ep_mxfp8_cutedsl_mega_multirank.py	20	CODE
LOW	tests/moe_ep/test_arch_and_build.py	3	CODE
LOW	…sts/moe_ep/test_moe_ep_nvfp4_cutedsl_mega_multirank.py	20	CODE
LOW	tests/moe_ep/test_layer_single_gpu.py	3	CODE
LOW	tests/moe_ep/smoke_nccl_ep.py	17	CODE
LOW	tests/moe_ep/test_mega_layer_validation.py	3	CODE
LOW	…s/moe_ep/test_mxfp8_cutedsl_preprocess_vs_reference.py	17	CODE
LOW	tests/moe_ep/nixl_ep/test_fleet_mock.py	17	CODE
LOW	tests/moe_ep/nccl_ep/conftest.py	9	CODE
LOW	tests/moe_ep/nccl_ep/test_handle_mock.py	14	CODE
LOW	tests/moe_ep/nccl_ep/test_fleet_mock.py	9	CODE
LOW	tests/moe_ep/nccl_ep/test_ndtensor.py	9	CODE
LOW	tests/comm/test_mnnvl_dcp_alltoall.py	64	CODE
LOW	tests/gdn/test_decode_pretranspose_bf16_padded_pool.py	29	CODE
LOW	tests/gdn/test_prefill_delta_rule.py	17	CODE
LOW	tests/gdn/test_decode_delta_rule.py	17	CODE
LOW	…sts/gdn/test_decode_pretranspose_noncontiguous_pool.py	17	CODE
LOW	tests/trace_apply/test_trace_apply.py	15	CODE
LOW	tests/moe/test_unified_moe.py	31	CODE
LOW	tests/moe/test_unified_moe_fuzz.py	140	CODE
LOW	flashinfer-jit-cache/flashinfer_jit_cache/__init__.py	17	CODE
LOW	flashinfer-jit-cache/flashinfer_jit_cache/__init__.py	19	CODE
LOW	flashinfer-jit-cache/flashinfer_jit_cache/__init__.py	19	CODE
LOW	benchmarks/bench_batch_attention.py	1	CODE
LOW	benchmarks/bench_ep_matrix.py	27	CODE
LOW	benchmarks/bench_moe_ep.py	49	CODE
LOW	flashinfer/fp8_quantization.py	12	CODE
LOW	flashinfer/fp8_quantization.py	12	CODE
LOW	flashinfer/fp8_quantization.py	12	CODE
LOW	flashinfer/fp8_quantization.py	12	CODE
LOW	flashinfer/__init__.py	18	CODE
LOW	flashinfer/__init__.py	20	CODE
LOW	flashinfer/__init__.py	21	CODE
LOW	flashinfer/__init__.py	24	CODE
LOW	flashinfer/__init__.py	25	CODE
LOW	flashinfer/__init__.py	26	CODE
LOW	flashinfer/__init__.py	27	CODE
LOW	flashinfer/__init__.py	28	CODE
LOW	flashinfer/__init__.py	31	CODE
LOW	flashinfer/__init__.py	32	CODE
LOW	flashinfer/__init__.py	35	CODE
LOW	flashinfer/__init__.py	36	CODE
1154 more matches not shown…

Over-Commented Block558 hits · 483 pts

Severity	File	Line	Snippet	Context
LOW	requirements.txt	1	apache-tvm-ffi>=0.1.6,!=0.1.8,!=0.1.8.post0,<0.2	COMMENT
LOW	.pre-commit-config.yaml	1	# To use:	COMMENT
LOW	pyproject.toml	1	# Copyright (c) 2024 by FlashInfer team.	COMMENT
LOW	build_backend.py	461		COMMENT
LOW	csrc/fmha_v2/fused_multihead_attention.h	201	int64_t softmax_stats_stride_in_bytes;	COMMENT
LOW	…/fmha_v2/fused_multihead_attention_kernel_4x1_hopper.h	461	// Create buffer_full barriers with 1 arrive count	COMMENT
LOW	csrc/fmha_v2/fused_multihead_attention_utils.h	801	int const* cu_mask_rows) {	COMMENT
LOW	…/fused_multihead_flash_attention_kernel_noloop_tiled.h	241	float global_sum[Softmax::ROWS_PER_THREAD];	COMMENT
LOW	csrc/fmha_v2/fused_multihead_attention_kernel.h	21	#include <fmha/smem_tile_o.h>	COMMENT
LOW	csrc/fmha_v2/fused_multihead_attention_kernel.h	41	// - P = norm * K^T * Q , where norm is the normalization term (a scalar),	COMMENT
LOW	…c/fmha_v2/fused_multihead_attention_demo_bert_params.h	101	// The stride between matrices of packed mask.	COMMENT
LOW	…c/fmha_v2/fused_multihead_attention_demo_bert_params.h	181	// Paged KV uses 4D tensor, the tensor size is:	COMMENT
LOW	csrc/fmha_v2/fmha/softmax.h	1041	}	COMMENT
LOW	csrc/fmha_v2/fmha/mask.h	341	}	COMMENT
LOW	csrc/fmha_v2/fmha/utils.h	21	#include <__clang_cuda_builtin_vars.h>	COMMENT
LOW	csrc/fmha_v2/fmha/utils.h	61	using Unpack_remaining = typename Unpack<Ns...>::Type;	COMMENT
LOW	csrc/fmha_v2/fmha/utils.h	81	// (84): here	COMMENT
LOW	csrc/fmha_v2/fmha/utils.h	101	struct Cond_inspect_ns {	COMMENT
LOW	csrc/fmha_v2/fmha/utils.h	801	// asm volatile("fma.rn.f16x2.relu %0, %1, %2, %3;" : "=r"(c) : "r"(a), "r"(one),	COMMENT
LOW	csrc/fmha_v2/fmha/utils.h	821	// return dst;	COMMENT
LOW	csrc/fmha_v2/fmha/paged_kv_cache.h	21	struct Kv_block_array {	COMMENT
LOW	csrc/fmha_v2/fmha/gmem_tile_o_packed.h	1281	// The scale.	COMMENT
LOW	csrc/fmha_v2/fmha/kernel_traits.h	21	#include <fmha/gmem_tile_qkv.h>	COMMENT
LOW	csrc/fmha_v2/fmha/fragment.h	141	f.reg(4 * ii + 0) = tmp.x;	COMMENT
LOW	csrc/fmha_v2/fmha/smem_tile_v.h	481	t[3] = y;	COMMENT
LOW	csrc/fmha_v2/fmha/smem_tile_v.h	601	// Jump by 8*16 rows per K but account for packing.	COMMENT
LOW	csrc/fmha_v2/fmha/smem_tile_v.h	921	uint2 tmp_0;	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/dma.h	301	if (q_step_idx == 0 && kv_step_idx == kv_idx_start) {	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/dma.h	681	// expected phase, while slow DMA warps are still waiting on the old expected phase of the	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/dma.h	701	//	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/compute.h	581	// results once. For example, STEP_KV=256, qgmma.K=32, then 256/32=8 MMAs are needs,	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/kernel_traits.h	41	int Q_BUFFERS_,	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/kernel_traits.h	201	// Apply the exp2f optimization (fuse bmm1_scale and -max into FMAs).	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/kernel_traits.h	421	// The step size in key/value sequence dimension (N of BMM1 and K of BMM2).	COMMENT
LOW	csrc/fmha_v2/fmha/hopper/arrive_wait.h	21		COMMENT
LOW	csrc/fmha_v2/fmha/hopper/arrive_wait.h	41	#endif	COMMENT
LOW	csrc/fmha_v2/fmha/hopper/smem_tile.h	1801	tmp[nii][ki].w); // PRMT 2+3	COMMENT
LOW	csrc/fmha_v2/fmha/hopper/smem_tile.h	1821	// [g0, g1] [o0, o1] => y: [g0, h0, o0, p0], i.e. col 0	COMMENT
LOW	csrc/fmha_v2/fmha/hopper/smem_tile.h	2001	// [00, 01]	COMMENT
LOW	csrc/fmha_v2/fmha/hopper/smem_tile.h	2121	BYTES_PER_STS, 1, 0, ROWS_PER_XOR_PATTERN, 1>;	COMMENT
LOW	csrc/fmha_v2/fmha/hopper/smem_tile.h	2321	// w: +24 (o)	COMMENT
LOW	csrc/fmha_v2/fmha/hopper/gmem_tile_o_packed.h	601	int64_t row_offset =	COMMENT
LOW	csrc/fmha_v2/fmha/hopper/fragment.h	281	Gmma_single_desc_b::TRANS_MODE == fmha::Gmma_descriptor_transpose::TRANS ? true : false,	COMMENT
LOW	csrc/fused_moe/monomoe/src/ptx_utils.h	1	#ifndef PTX_UTILS_H	COMMENT
LOW	csrc/fused_moe/monomoe/src/ptx_utils.h	21	// * Issued by a full warpgroup (4 consecutive warps = 128 threads).	COMMENT
LOW	csrc/fused_moe/monomoe/src/ptx_utils.h	181	// 1. Arrival counter — decremented by `mbarrier.arrive*`.	COMMENT
LOW	csrc/fused_moe/monomoe/src/ptx_utils.h	321	// ── Hopper TMA bulk-tensor copy (sm_90a) ──────────────────────────────────	COMMENT
LOW	csrc/fused_moe/monomoe/src/ptx_utils.h	341	// covers every TMA issue pointing at this same barrier, before	COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_tma.h	1	#ifndef MOE_TMA_H	COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_tma.h	21	// CUDA Driver API: provides `CUtensorMap` and `cuTensorMapEncodeTiled`.	COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_tma.h	441	// `create_down_weight_tma_desc` and `create_down_activation_tma_desc`.	COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_grid_barrier.h	1	#pragma once	COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_internal.h	1		COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_internal.h	21	// pipeline / data-movement reschedules — they do not change the kernel's	COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_internal.h	101		COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_internal.h	221	static constexpr uint32_t value = Dims::DOWN_SCALE_COLS;	COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_internal.h	261		COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_internal.h	321	"down-projection (one outer K-step consumes K_STEP_DOWN "	COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_internal.h	341	static_assert(!use_wgmma_v<Dims> \|\| Dims::HIDDEN_STATES % K_STEP_UP == 0,	COMMENT
LOW	csrc/fused_moe/monomoe/src/moe_internal.h	361	// the single-buffer `spec->down_partial_out[BS][HIDDEN_STATES]`.	COMMENT
498 more matches not shown…

Deep Nesting409 hits · 355 pts

Severity	File	Line	Context
LOW	tests/conftest.py	175	CODE
LOW	tests/conftest.py	221	CODE
LOW	tests/attention/test_trtllm_gen_sparse_mla_dsv4.py	194	CODE
LOW	tests/attention/test_fmha_v2_prefill.py	62	CODE
LOW	tests/attention/test_fmha_v2_prefill.py	208	CODE
LOW	tests/attention/test_fmha_v2_prefill.py	464	CODE
LOW	tests/attention/test_fmha_v2_prefill.py	1177	CODE
LOW	tests/attention/test_batch_prefill_kernels.py	76	CODE
LOW	tests/attention/test_batch_prefill_kernels.py	416	CODE
LOW	tests/attention/test_attention_sink.py	36	CODE
LOW	tests/attention/test_xqa.py	164	CODE
LOW	tests/attention/test_cute_dsl_mla_decode.py	758	CODE
LOW	tests/attention/test_rope.py	1392	CODE
LOW	tests/attention/test_vsa_block_sparse.py	264	CODE
LOW	tests/attention/test_vsa_block_sparse.py	919	CODE
LOW	tests/attention/test_vsa_block_sparse.py	1230	CODE
LOW	tests/trace/test_fi_trace_template_consistency.py	191	CODE
LOW	tests/mamba/conftest.py	720	CODE
LOW	tests/mamba/triton_reference/ssd_chunk_scan.py	88	CODE
LOW	…amba/triton_reference/selective_state_update_varlen.py	272	CODE
LOW	tests/mamba/triton_reference/ssd_chunk_state.py	387	CODE
LOW	tests/mamba/triton_reference/selective_state_update.py	81	CODE
LOW	tests/mamba/triton_reference/ssd_state_passing.py	36	CODE
LOW	…s/mamba/triton_reference/checkpointing_state_update.py	436	CODE
LOW	…s/mamba/triton_reference/checkpointing_state_update.py	925	CODE
LOW	tests/comm/test_trtllm_alltoall.py	536	CODE
LOW	tests/comm/test_trtllm_alltoall.py	584	CODE
LOW	tests/comm/test_trtllm_allreduce_fusion.py	27	CODE
LOW	tests/comm/test_trtllm_moe_alltoall.py	703	CODE
LOW	tests/comm/test_trtllm_allreduce.py	31	CODE
LOW	tests/comm/test_allreduce_fusion_moe_unified_api.py	79	CODE
LOW	tests/comm/test_trtllm_moe_allreduce_fusion_finalize.py	25	CODE
LOW	tests/comm/test_mnnvl_moe_alltoall.py	124	CODE
LOW	tests/comm/test_mnnvl_moe_alltoall.py	440	CODE
LOW	tests/comm/test_trtllm_mnnvl_allreduce.py	793	CODE
LOW	…s/comm/test_trtllm_allreduce_fusion_group_fp8_quant.py	297	CODE
LOW	tests/comm/test_mixed_comm.py	46	CODE
LOW	tests/comm/test_mixed_comm.py	204	CODE
LOW	tests/comm/test_trtllm_moe_allreduce_fusion.py	25	CODE
LOW	tests/comm/test_vllm_custom_allreduce.py	35	CODE
LOW	tests/utils/test_fp4_quantize.py	808	CODE
LOW	tests/utils/test_gen_module_symlink_race_condition.py	66	CODE
LOW	tests/utils/test_logging_replay.py	366	CODE
LOW	tests/utils/test_logging_replay.py	440	CODE
LOW	tests/utils/test_logging_replay.py	514	CODE
LOW	tests/utils/test_logging_replay.py	587	CODE
LOW	tests/utils/test_logging_replay.py	644	CODE
LOW	tests/utils/test_logging_replay.py	721	CODE
LOW	tests/utils/test_logging_replay.py	825	CODE
LOW	tests/utils/test_logging_replay.py	912	CODE
LOW	tests/utils/test_sampling.py	773	CODE
LOW	tests/utils/test_sampling.py	1064	CODE
LOW	tests/utils/test_topk.py	1083	CODE
LOW	tests/model_optimizations/test_dsv3_fused_routing.py	133	CODE
LOW	tests/test_helpers/sink_attention_reference.py	39	CODE
LOW	tests/moe/test_trtllm_cutlass_fused_moe.py	312	CODE
LOW	tests/moe/trtllm_gen_fused_moe_utils.py	2480	CODE
LOW	tests/moe/trtllm_gen_fused_moe_utils.py	3275	CODE
LOW	tests/moe/trtllm_gen_fused_moe_utils.py	1110	CODE
LOW	tests/moe/test_b12x_fused_moe.py	665	CODE
349 more matches not shown…

Excessive Try-Catch Wrapping255 hits · 245 pts

Severity	File	Line	Snippet	Context
LOW	build_backend.py	131	except Exception:	CODE
LOW	build_backend.py	193	except Exception:	CODE
LOW	build_backend.py	208	except Exception:	CODE
LOW	build_backend.py	313	except Exception:	CODE
LOW	build_backend.py	390	except Exception:	CODE
LOW	build_backend.py	883	except Exception as e:	CODE
LOW	build_utils.py	45	except Exception:	CODE
LOW	tests/attention/test_deepseek_mla.py	102	except Exception as e:	CODE
MEDIUM	tests/attention/test_deepseek_mla.py	42	def warmup_jit():	CODE
LOW	…sts/trace/test_tgv_gemm_sm100_reference_correctness.py	39	except Exception as exc:	CODE
LOW	…sts/trace/test_mxfp8_quantize_reference_correctness.py	25	except Exception as exc:	CODE
LOW	tests/trace/test_mm_bf16_fp4_reference_correctness.py	44	except Exception as exc:	CODE
LOW	tests/trace/test_single_decode_reference_correctness.py	31	except Exception as exc:	CODE
LOW	…ize_fp8_append_paged_kv_cache_reference_correctness.py	69	except Exception as exc:	CODE
LOW	tests/trace/test_bmm_fp8_reference_correctness.py	38	except Exception as exc:	CODE
LOW	…/trace/test_cutlass_fused_moe_reference_correctness.py	62	except Exception as exc:	CODE
LOW	…s/trace/test_segment_gemm_run_reference_correctness.py	34	except Exception as exc:	CODE
LOW	…s/trace/test_block_sparse_run_reference_correctness.py	58	except Exception as exc:	CODE
LOW	…s/trace/test_xqa_batch_decode_reference_correctness.py	74	except Exception as exc:	CODE
LOW	…sts/trace/test_mxfp4_quantize_reference_correctness.py	33	except Exception as exc:	CODE
LOW	…ace/test_var_block_sparse_run_reference_correctness.py	67	except Exception as exc:	CODE
LOW	tests/trace/test_bmm_bf16_reference_correctness.py	29	except Exception as exc:	CODE
LOW	tests/trace/test_mm_bf16_reference_correctness.py	29	except Exception as exc:	CODE
LOW	tests/trace/test_fi_trace_template_consistency.py	260	except Exception as exc: # noqa: BLE001	CODE
LOW	…ace/test_xqa_batch_decode_mla_reference_correctness.py	72	except Exception as exc:	CODE
LOW	…trace/test_cudnn_batch_decode_reference_correctness.py	80	except Exception as exc:	CODE
LOW⚡	tests/trace/example.py	389	except Exception:	CODE
LOW⚡	tests/trace/example.py	400	except Exception:	CODE
LOW	tests/trace/example.py	414	except Exception:	CODE
LOW	tests/trace/example.py	426	except Exception:	CODE
LOW	tests/trace/example.py	798	except Exception:	CODE
LOW	…race/test_batch_attention_run_reference_correctness.py	61	except Exception as exc:	CODE
LOW	…e/test_trtllm_fmha_v2_prefill_reference_correctness.py	62	except Exception as exc:	CODE
LOW	tests/trace/test_concat_mla_k_reference_correctness.py	30	except Exception as exc:	CODE
LOW	…/test_trtllm_batch_decode_mla_reference_correctness.py	68	except Exception as exc:	CODE
LOW	…sts/trace/test_single_prefill_reference_correctness.py	31	except Exception as exc:	CODE
LOW	…/test_multi_level_cascade_run_reference_correctness.py	61	except Exception as exc:	CODE
LOW	…race/test_cudnn_batch_prefill_reference_correctness.py	84	except Exception as exc:	CODE
LOW	…/test_fused_add_rmsnorm_quant_reference_correctness.py	32	except Exception as exc:	CODE
LOW	…t_pod_with_paged_kv_cache_run_reference_correctness.py	70	except Exception as exc:	CODE
LOW	tests/trace/test_rmsnorm_quant_reference_correctness.py	27	except Exception as exc:	CODE
LOW	…sts/trace/test_nvfp4_quantize_reference_correctness.py	26	except Exception as exc:	CODE
LOW	tests/trace/test_batch_pod_run_reference_correctness.py	76	except Exception as exc:	CODE
LOW	tests/autotuner/test_autotuner_distributed.py	250	except Exception as e:	CODE
MEDIUM	tests/autotuner/test_autotuner_configs.py	888	def do_search():	CODE
MEDIUM	tests/autotuner/test_autotuner_configs.py	926	def do_load():	CODE
MEDIUM	tests/autotuner/test_autotuner_configs.py	934	def do_search():	CODE
MEDIUM	tests/autotuner/test_autotuner_configs.py	967	def run_autotune():	CODE
LOW	tests/autotuner/test_autotuner_configs.py	897	except Exception as e:	CODE
LOW	tests/autotuner/test_autotuner_configs.py	931	except Exception as e:	CODE
LOW	tests/autotuner/test_autotuner_configs.py	939	except Exception as e:	CODE
LOW	tests/autotuner/test_autotuner_configs.py	973	except Exception as e:	CODE
LOW	tests/mamba/conftest.py	748	except Exception as exc: # noqa: BLE001 — best-effort prewarm	CODE
LOW	tests/mamba/conftest.py	781	except Exception as exc: # noqa: BLE001 — best-effort prewarm	CODE
LOW	tests/comm/conftest.py	70	except Exception:	CODE
LOW	tests/comm/test_allreduce_unified_api.py	260	except Exception as e:	CODE
LOW	tests/comm/test_nvshmem_allreduce.py	59	except Exception as e:	CODE
LOW	tests/comm/test_trtllm_mnnvl_allreduce_custom_comm.py	197	except Exception as e:	CODE
MEDIUM	tests/comm/test_trtllm_mnnvl_allreduce_custom_comm.py	70	def get_open_port() -> int:	CODE
LOW	tests/comm/test_dcp_alltoall.py	48	except Exception:	CODE
195 more matches not shown…

AI Structural Patterns250 hits · 204 pts

Severity	File	Line	Context
LOW	tests/attention/test_fmha_v2_prefill.py	62	CODE
LOW	tests/attention/test_mla_decode_kernel.py	177	CODE
LOW	tests/attention/test_mla_decode_kernel.py	398	CODE
LOW	tests/attention/test_xqa.py	120	CODE
LOW	tests/attention/test_trtllm_gen_attention_decode.py	615	CODE
LOW	tests/attention/test_batch_attention.py	91	CODE
LOW	tests/mamba/test_chunk_scan_combined.py	27	CODE
LOW	tests/mamba/utils.py	16	CODE
LOW	…amba/triton_reference/selective_state_update_varlen.py	272	CODE
LOW	tests/mamba/triton_reference/selective_state_update.py	431	CODE
LOW	tests/mamba/triton_reference/ssd_combined.py	32	CODE
LOW	tests/mamba/triton_reference/ssd_combined.py	197	CODE
LOW	…s/mamba/triton_reference/checkpointing_state_update.py	925	CODE
LOW	tests/comm/test_all_gather_matmul.py	45	CODE
LOW	tests/comm/test_mixed_comm.py	43	CODE
LOW	tests/gemm/test_cute_dsl_gemm_allreduce_two_shot.py	139	CODE
LOW	tests/gemm/test_groupwise_scaled_gemm_mxfp4.py	73	CODE
LOW	tests/gdn/reference_delta_rule.py	125	CODE
LOW	tests/test_helpers/utils_fp4.py	51	CODE
LOW	tests/moe/test_dpsk_fused_moe_fp8.py	279	CODE
LOW	tests/moe/test_dpsk_fused_moe_fp8.py	370	CODE
LOW	tests/moe/trtllm_gen_fused_moe_utils.py	2291	CODE
LOW	tests/moe/trtllm_gen_fused_moe_utils.py	2439	CODE
LOW	tests/moe/trtllm_gen_fused_moe_utils.py	3275	CODE
LOW	tests/moe/trtllm_gen_fused_moe_utils.py	238	CODE
LOW	tests/moe/trtllm_gen_fused_moe_utils.py	2818	CODE
LOW	docs/tutorials/jax_tvm_ffi/gemma3_flashinfer_jax.py	834	CODE
LOW	flashinfer-jit-cache/build_backend.py	237	CODE
LOW	examples/pytorch/wan/transformer_wan_flashinfer.py	294	CODE
LOW	examples/pytorch/wan/transformer_wan_flashinfer.py	509	CODE
LOW	benchmarks/bench_gdn_decode.py	337	CODE
LOW	benchmarks/bench_gdn_decode.py	1046	CODE
LOW	benchmarks/bench_gdn_decode.py	1477	CODE
LOW	benchmarks/bench_moe_deepseek.py	198	CODE
LOW	benchmarks/bench_moe_deepseek.py	660	CODE
LOW	scripts/codeowner_analyzer.py	22	CODE
LOW	flashinfer/xqa.py	155	CODE
LOW	flashinfer/gdn_prefill.py	103	CODE
LOW	flashinfer/tactics_blocklist_gen.py	511	CODE
LOW	flashinfer/pod.py	269	CODE
LOW	flashinfer/pod.py	453	CODE
LOW	flashinfer/pod.py	906	CODE
LOW	flashinfer/pod.py	1127	CODE
LOW	flashinfer/kda_decode.py	41	CODE
LOW	flashinfer/rope.py	1286	CODE
LOW	flashinfer/rope.py	1364	CODE
LOW	flashinfer/rope.py	1504	CODE
LOW	flashinfer/sampling.py	1579	CODE
LOW	flashinfer/prefill.py	1130	CODE
LOW	flashinfer/prefill.py	1158	CODE
LOW	flashinfer/prefill.py	1186	CODE
LOW	flashinfer/prefill.py	4061	CODE
LOW	flashinfer/prefill.py	4082	CODE
LOW	flashinfer/prefill.py	4102	CODE
LOW	flashinfer/prefill.py	4478	CODE
LOW	flashinfer/prefill.py	4940	CODE
LOW	flashinfer/prefill.py	224	CODE
LOW	flashinfer/prefill.py	661	CODE
LOW	flashinfer/prefill.py	843	CODE
LOW	flashinfer/prefill.py	1608	CODE
190 more matches not shown…

Structural Annotation Overuse117 hits · 180 pts

Severity	File	Line	Snippet	Context
LOW	tests/attention/test_rope.py	863	# Step 1: Pre-populate cache with existing tokens	COMMENT
LOW	tests/attention/test_rope.py	1074	# Step 2: Append new tokens to the pre-populated cache	COMMENT
LOW	…_quantize_fp8_append_paged_kv_cache_h8_kv2_rope64.json	194	"reference": "from __future__ import annotations\nimport math\nimport torch\nimport torch.nn.functional as F\n\n@torch	CODE
LOW	…tize_fp8_append_paged_kv_cache_h8_kv2_rope64_ps16.json	195	"reference": "from __future__ import annotations\nimport math\nimport torch\nimport torch.nn.functional as F\n\n@torch	CODE
LOW⚡	tests/utils/test_fp4_quantize.py	309	# Step 1: Quantize with fp4_quantize	COMMENT
LOW⚡	tests/utils/test_fp4_quantize.py	319	# Step 2: Dequantize with e2m1_and_ufp8sf_scale_to_float	COMMENT
LOW	tests/utils/test_fp4_quantize.py	334	# Step 3: Compare results	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1081	# Step 1: Compute g = -exp(A_log) * softplus(a + dt_bias)	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1136	# Step 1: Apply gating to hidden state: h *= exp(g)	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1139	# Step 2: Delta rule: v -= k^T @ h ([B,H,K] x [B,H,K,V] -> [B,H,V])	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1142	# Step 3: Apply beta gating: v *= beta	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1145	# Step 4: Update hidden state with outer product: h += k ⊗ v_new	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1148	# Step 5: Compute output: o = q^T @ h	COMMENT
LOW	tests/gdn/reference_delta_rule.py	1098	# Step 2: Compute beta = sigmoid(b)	COMMENT
LOW	.claude/skills/benchmark-kernel/SKILL.md	53	### Step 1: Choose Your Test Routine	COMMENT
LOW	.claude/skills/benchmark-kernel/SKILL.md	60	### Step 2: Run a Single Benchmark	COMMENT
LOW	.claude/skills/benchmark-kernel/SKILL.md	108	### Step 3: Understand the Output	COMMENT
LOW	.claude/skills/benchmark-kernel/SKILL.md	124	### Step 4: Run Batch Benchmarks	COMMENT
LOW	.claude/skills/benchmark-kernel/SKILL.md	146	### Step 5: Common Flags	COMMENT
LOW	.claude/skills/benchmark-kernel/SKILL.md	164	### Step 1: Write Your Benchmark Script	COMMENT
LOW	.claude/skills/benchmark-kernel/SKILL.md	200	### Step 2: Run Your Benchmark	COMMENT
LOW	.claude/skills/benchmark-kernel/SKILL.md	219	### Step 3: Advanced Options	COMMENT
LOW	.claude/skills/debug-cuda-crash/SKILL.md	24	## Step 1: Enable API Logging	COMMENT
LOW	.claude/skills/debug-cuda-crash/SKILL.md	109	## Step 2: Reproduce the Crash	COMMENT
LOW	.claude/skills/debug-cuda-crash/SKILL.md	146	## Step 3: Common CUDA Errors and How to Debug	COMMENT
LOW	.claude/skills/debug-cuda-crash/SKILL.md	249	## Step 4: Multi-Process Debugging	COMMENT
LOW	.claude/skills/debug-cuda-crash/SKILL.md	268	## Step 5: Advanced Debugging with compute-sanitizer	COMMENT
LOW	.claude/skills/debug-cuda-crash/SKILL.md	309	## Step 6: Kernel-Level Debugging with printf()	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	18	## Step 1: Define CUDA Kernel in `include/`	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	76	## Step 2: Create Launcher in `csrc/`	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	163	## Step 3: Create TVM-FFI Binding in `csrc/`	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	183	## Step 4: Create JIT Generator (No Jinja for Simple Case)	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	387	## Step 5: Create Python API in `flashinfer/`	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	628	## Step 6: Add a Trace Template	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	776	## Step 7: Write Tests in `tests/`	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	846	## Step 7: Register in AOT	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	872	## Step 8: Export API	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	883	## Step 9: Run and Test	COMMENT
LOW	.claude/skills/add-cuda-kernel/SKILL.md	893	## Step 10: Add Benchmark	COMMENT
LOW	docs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py	176	# Step 1 - Compile and load the TVM FFI module	COMMENT
LOW	docs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py	197	# Step 2 - Register as a JAX FFI target	COMMENT
LOW	docs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py	240	# Step 3 - Call as a regular JAX expression	COMMENT
LOW⚡	benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py	149	# Step 1: Add	COMMENT
LOW⚡	benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py	151	# Step 2: RMSNorm	COMMENT
LOW⚡	benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py	153	# Step 3: FP4 quantize (with global_scale for NVFP4)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	179	# Step 1: Apply decay to hidden state: h *= exp(g)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	182	# Step 2: Delta rule: v -= sum(h * k, dim=0) = k @ h	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	187	# Step 3: Apply beta gating: v *= beta	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	190	# Step 4: Update hidden state: h += outer(k, v) = k[:, None] * v[None, :]	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	193	# Step 5: Compute output: o = q @ h = sum(q[:, None] * h, dim=0)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	361	# Step 1: Apply decay: h *= exp(g)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	364	# Step 2: Delta rule: v -= h @ k (h is [BV, BK], k is [BK])	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	369	# Step 3: Apply beta gating	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	372	# Step 4: Update state: h += outer(v, k) = v[:, None] * k[None, :]	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	376	# Step 5: Compute output: o = h @ q = sum(h * q[None, :], axis=1) -> [BV]	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	633	# Step 1: Apply decay to hidden state: h *= exp(g)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	636	# Step 2: Delta rule: v -= h @ k = sum(h * k[None, :], axis=1)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	640	# Step 3: Apply beta gating: v *= beta	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	643	# Step 4: Update hidden state: h += outer(v, k) = v[:, None] * k[None, :]	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	646	# Step 5: Compute output: o = h @ q = sum(h * q[None, :], axis=1)	COMMENT
57 more matches not shown…

Self-Referential Comments55 hits · 147 pts

Severity	File	Line	Snippet	Context
MEDIUM	tests/conftest.py	234	# Create a dict with module info for reporting	COMMENT
MEDIUM	tests/attention/test_attention_sink.py	287	# Create a fragmented page allocation pattern	COMMENT
MEDIUM	tests/attention/test_trtllm_gen_attention_decode.py	1562	# Create a larger tensor with 2*D in the last dimension	COMMENT
MEDIUM⚡	tests/trace/test_fi_trace.py	576	# Create a dummy instance — we don't call run(), only fi_trace()	COMMENT
MEDIUM	tests/autotuner/test_autotuner_configs.py	725	# Create a cache file with a known config	COMMENT
MEDIUM	tests/autotuner/test_autotuner_configs.py	916	# Create a cache file	COMMENT
MEDIUM	tests/mamba/utils.py	266	# Create a simple linear chain structure by default	COMMENT
MEDIUM	tests/comm/test_trtllm_alltoall.py	114	# Create a random input tensor	COMMENT
MEDIUM	tests/comm/test_trtllm_alltoall.py	177	# Create a random input tensor	COMMENT
MEDIUM	tests/comm/test_trtllm_moe_alltoall.py	265	# Create a random input tensor	COMMENT
MEDIUM	tests/comm/test_trtllm_moe_alltoall.py	506	# Create a random input tensor	COMMENT
MEDIUM	tests/comm/test_trtllm_moe_alltoall.py	745	# Create a random input tensor	COMMENT
MEDIUM	tests/comm/test_mnnvl_memory.py	135	# Create a random input tensor	COMMENT
MEDIUM	tests/utils/test_fp4_quantize.py	247	# Create a test scale factors tensor with uint8 dtype	COMMENT
MEDIUM	tests/utils/test_fp4_quantize.py	297	# Create a reasonable test tensor	COMMENT
MEDIUM	tests/utils/test_logging.py	400	# Create a tensor with specific properties	COMMENT
MEDIUM	tests/utils/test_logging_replay.py	1016	# Create a mock dump with safetensors files	COMMENT
MEDIUM	tests/utils/test_norm.py	719	# Create a JIT spec for norm module without ENABLE_FP8 flag	COMMENT
MEDIUM	tests/utils/test_load_cubin_compile_race_condition.py	40	# Define the target file - same for all processes	COMMENT
MEDIUM	tests/utils/test_load_cubin_compile_race_condition.py	75	# Create a temporary directory for FLASHINFER_CUBIN_DIR	COMMENT
MEDIUM	tests/cli/test_cli_cmds.py	139	# Create a temporary JIT directory with some dummy cache files	COMMENT
MEDIUM	tests/cli/test_cli_cmds.py	182	# Create a temporary cubin directory with some dummy cubin files	COMMENT
MEDIUM	tests/test_helpers/rope_reference.py	113	# The following code is from the vLLM's implementation of RoPE.	COMMENT
MEDIUM	tests/moe/test_dpsk_fused_moe_fp8.py	596	# Create a mock MoE implementation for skip_checks	COMMENT
MEDIUM	docs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py	1	# This file is the canonical Sphinx-Gallery source for the tutorial.	COMMENT
MEDIUM	docs/tutorials/jax_tvm_ffi/gemma3_flashinfer_jax.py	1	# This file is the canonical Sphinx-Gallery source for the tutorial.	COMMENT
MEDIUM	examples/pytorch/wan/transformer_wan_flashinfer.py	4	# This file is adapted from the diffusers WanTransformer3DModel implementation:	COMMENT
MEDIUM	scripts/task_test_multi_gpu_comm_kernels.sh	22	# Define the specific test files for multi-GPU comm tests (single-node)	COMMENT
MEDIUM	scripts/task_test_multi_node_comm_kernels.sh	25	# Define the specific test files for multi-node comm tests	COMMENT
MEDIUM	scripts/test_utils.sh	3	# This file is meant to be sourced by test runner scripts	COMMENT
MEDIUM	scripts/test_utils.sh	975	# Create a bash array with the node IDs	COMMENT
MEDIUM	scripts/test_utils.sh	1145	# Create a temporary directory for parallel job state	COMMENT
MEDIUM	scripts/test_utils.sh	1160	# Create a results file for each test	COMMENT
MEDIUM	…/cutedsl_backend_kernels/moe_nvfp4_swapab/moe_utils.py	219	# Create a new cute.Pointer with generic address space, preserving alignment	COMMENT
MEDIUM	…/cutedsl_backend_kernels/moe_nvfp4_swapab/moe_utils.py	247	# Create a new cute.Pointer with gmem address space, preserving alignment	COMMENT
MEDIUM	flashinfer/comm/dlpack_utils.py	224	# Create a new capsule each time	STRING
MEDIUM	flashinfer/fused_moe/core.py	152	# Create a unique cache key (weight_type, weight_shape)	COMMENT
MEDIUM	flashinfer/fused_moe/core.py	184	# Create a unique cache key (weight_type, weight_shape)	COMMENT
MEDIUM	flashinfer/fused_moe/core.py	221	# Create the output	COMMENT
MEDIUM	…hinfer/fused_moe/cute_dsl/blackwell/custom_pipeline.py	44	# This file is copied and modified from https://github.com/NVIDIA/cutlass/tree/main/python/CuTeDSL/cutlass/pipeline	COMMENT
MEDIUM	flashinfer/fused_moe/cute_dsl/blackwell/utils.py	44	# This file is copied and modified from cutlass https://github.com/NVIDIA/cutlass/blob/main/python/CuTeDSL/cutlass/cute/	COMMENT
MEDIUM	flashinfer/fused_moe/cute_dsl/blackwell/utils.py	174	# Create a numpy array	STRING
MEDIUM	…blockscaled_contiguous_grouped_gemm_finalize_fusion.py	315	"""This class implements batched matrix multiplication (C = A x SFA x B x SFB) with support for various data types	STRING
MEDIUM	…ockscaled_contiguous_gather_grouped_gemm_act_fusion.py	322	"""This class implements contiguous grouped matrix multiplication with gather operation and SwiGLU fusion	STRING
MEDIUM	flashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py	29	# This file is ported from CUTLASS's sm103_dense_blockscaled_gemm_persistent.py	COMMENT
MEDIUM	flashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py	50	"""This class implements batched matrix multiplication (C = A x SFA x B x SFB) with support for FP4 data types	STRING
MEDIUM	…ashinfer/gemm/kernels/grouped_gemm_masked_blackwell.py	586	"""This class implements batched matrix multiplication (C = A x SFA x B x SFB) with support for various data types	STRING
MEDIUM	…nfer/gemm/kernels/dense_blockscaled_gemm_sm120_b12x.py	29	# This file is ported from the CUTLASS dense block-scaled GEMM example	COMMENT
MEDIUM	flashinfer/gemm/kernels/dense_blockscaled_gemm_sm100.py	29	# This file is ported from TensorRT-LLM's dense_blockscaled_gemm_persistent.py	COMMENT
MEDIUM	flashinfer/cute_dsl/gemm_allreduce_two_shot.py	217	"""This class implements batched matrix multiplication (C = A x B) with support for various data types	STRING
MEDIUM	flashinfer/cute_dsl/utils.py	330	# Create a numpy array	STRING
MEDIUM	flashinfer/jit/cubin_loader.py	311	# Define the correct callback type	COMMENT
MEDIUM	flashinfer/jit/cubin_loader.py	321	# Create the callback and keep a reference to prevent GC	COMMENT
MEDIUM	flashinfer/jit/gemm/cutlass/cutlass_library.py	40	# The following block implements enum.auto() for Python 3.5 variants that don't include it such	COMMENT
MEDIUM	flashinfer/logits_processor/processors.py	42	>>> # Create a pipeline that legalizes to a fused op.	STRING

Verbosity Indicators94 hits · 147 pts

Severity	File	Line	Snippet	Context
LOW	…mha_v2/fused_multihead_flash_attention_kernel_noloop.h	286	// Do we need to check if there are negative inf for softmax row_max ?	COMMENT
LOW	…/fused_multihead_flash_attention_kernel_noloop_tiled.h	92	// Do we need to check if there are negative inf for softmax row_max ?	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/epilogue.h	73	// Whether we need to check if local_max could be -inf or not.	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/epilogue.h	543	// Whether we need to check if local_max could be -inf or not.	COMMENT
LOW	csrc/fmha_v2/fmha/warpspec/epilogue.h	734	// Whether we need to check if local_max could be -inf or not.	COMMENT
LOW	tests/attention/test_rope.py	863	# Step 1: Pre-populate cache with existing tokens	COMMENT
LOW	tests/attention/test_rope.py	1074	# Step 2: Append new tokens to the pre-populated cache	COMMENT
LOW	…_quantize_fp8_append_paged_kv_cache_h8_kv2_rope64.json	194	"reference": "from __future__ import annotations\nimport math\nimport torch\nimport torch.nn.functional as F\n\n@torch	CODE
LOW	…tize_fp8_append_paged_kv_cache_h8_kv2_rope64_ps16.json	195	"reference": "from __future__ import annotations\nimport math\nimport torch\nimport torch.nn.functional as F\n\n@torch	CODE
LOW⚡	tests/utils/test_fp4_quantize.py	309	# Step 1: Quantize with fp4_quantize	COMMENT
LOW⚡	tests/utils/test_fp4_quantize.py	319	# Step 2: Dequantize with e2m1_and_ufp8sf_scale_to_float	COMMENT
LOW	tests/utils/test_fp4_quantize.py	334	# Step 3: Compare results	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1081	# Step 1: Compute g = -exp(A_log) * softplus(a + dt_bias)	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1136	# Step 1: Apply gating to hidden state: h *= exp(g)	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1139	# Step 2: Delta rule: v -= k^T @ h ([B,H,K] x [B,H,K,V] -> [B,H,V])	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1142	# Step 3: Apply beta gating: v *= beta	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1145	# Step 4: Update hidden state with outer product: h += k ⊗ v_new	COMMENT
LOW⚡	tests/gdn/reference_delta_rule.py	1148	# Step 5: Compute output: o = q^T @ h	COMMENT
LOW	tests/gdn/reference_delta_rule.py	1098	# Step 2: Compute beta = sigmoid(b)	COMMENT
LOW⚡	benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py	149	# Step 1: Add	COMMENT
LOW⚡	benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py	151	# Step 2: RMSNorm	COMMENT
LOW⚡	benchmarks/bench_cute_dsl_add_rmsnorm_fp4quant.py	153	# Step 3: FP4 quantize (with global_scale for NVFP4)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	179	# Step 1: Apply decay to hidden state: h *= exp(g)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	182	# Step 2: Delta rule: v -= sum(h * k, dim=0) = k @ h	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	187	# Step 3: Apply beta gating: v *= beta	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	190	# Step 4: Update hidden state: h += outer(k, v) = k[:, None] * v[None, :]	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	193	# Step 5: Compute output: o = q @ h = sum(q[:, None] * h, dim=0)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	361	# Step 1: Apply decay: h *= exp(g)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	364	# Step 2: Delta rule: v -= h @ k (h is [BV, BK], k is [BK])	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	369	# Step 3: Apply beta gating	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	372	# Step 4: Update state: h += outer(v, k) = v[:, None] * k[None, :]	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	376	# Step 5: Compute output: o = h @ q = sum(h * q[None, :], axis=1) -> [BV]	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	633	# Step 1: Apply decay to hidden state: h *= exp(g)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	636	# Step 2: Delta rule: v -= h @ k = sum(h * k[None, :], axis=1)	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	640	# Step 3: Apply beta gating: v *= beta	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	643	# Step 4: Update hidden state: h += outer(v, k) = v[:, None] * k[None, :]	COMMENT
LOW⚡	benchmarks/gdn_triton_reference.py	646	# Step 5: Compute output: o = h @ q = sum(h * q[None, :], axis=1)	COMMENT
LOW⚡	benchmarks/bench_cute_dsl_rmsnorm_fp4quant.py	157	# Step 1: RMSNorm	COMMENT
LOW⚡	benchmarks/bench_cute_dsl_rmsnorm_fp4quant.py	159	# Step 2: FP4 quantize (with global_scale for NVFP4)	COMMENT
LOW	benchmarks/routines/norm.py	1120	# Step 2: RMSNorm on residual	COMMENT
LOW	flashinfer/trace/templates/rope.py	1018	# Step 3: append into paged cache (GQA/MHA) — materialize the quantized	COMMENT
LOW	…l/cutedsl_backend_kernels/moe_nvfp4_swapab/epilogue.py	889	# Step 1: absmax over the weighted regs.	COMMENT
LOW	…_backend_kernels/moe_nvfp4_swapab/epilogue_refactor.py	1062	# Step 0: load tmem	COMMENT
LOW	…_backend_kernels/moe_nvfp4_swapab/epilogue_refactor.py	1116	# Step 1: perform swiglu on the first part, interleave with the second's 32x32 tmem transpose.	COMMENT
LOW	…_backend_kernels/moe_nvfp4_swapab/epilogue_refactor.py	1148	# Step 2: Quant	COMMENT
LOW	…_backend_kernels/moe_nvfp4_swapab/epilogue_refactor.py	1159	# Step 3: TMASTG	COMMENT
LOW⚡	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	911	# Step 1: deterministic input + weight generation	COMMENT
LOW⚡	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	1265	# Step 2: reference	COMMENT
LOW⚡	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	1297	# Step 3: workspace allocation	COMMENT
LOW⚡	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	1338	# Step 4: kernel launch	COMMENT
LOW⚡	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	1910	# Step 5: validation	COMMENT
LOW⚡	…utedsl_backend_kernels/moe_nvfp4_swapab/kernel_fc12.py	1884	# Step 1: coarse-grain spin on ``fc1_done_counter[slot]``	COMMENT
LOW	…utedsl_backend_kernels/moe_nvfp4_swapab/kernel_fc12.py	1979	# Step 3: K-loop with 2x cute.copy per tile (B +	COMMENT
LOW⚡	…l/cutedsl_backend_kernels/moe_mxfp8_glu/mega_runner.py	221	# Step 1: deterministic input + weight generation (MXFP8)	COMMENT
LOW⚡	…l/cutedsl_backend_kernels/moe_mxfp8_glu/mega_runner.py	403	# Step 2: reference (MXFP8)	COMMENT
LOW⚡	…l/cutedsl_backend_kernels/moe_mxfp8_glu/mega_runner.py	430	# Step 5: validation (MXFP8)	COMMENT
LOW⚡	…l/cutedsl_backend_kernels/moe_mxfp8_glu/mega_runner.py	470	# Step 4: kernel launch (MXFP8)	COMMENT
LOW⚡	flashinfer/gdn_kernels/gdn_decode_pretranspose.py	278	# Step 1: Wait for current stage to complete	COMMENT
LOW⚡	flashinfer/gdn_kernels/gdn_decode_pretranspose.py	282	# Step 2: Issue async load for next tile (after compute)	COMMENT
LOW⚡	flashinfer/gdn_kernels/gdn_decode_pretranspose.py	589	# Step 1: Wait for current stage to complete	COMMENT
34 more matches not shown…

Modern Structural Boilerplate137 hits · 130 pts

Severity	File	Line	Snippet	Context
LOW	flashinfer-cubin/flashinfer_cubin/__init__.py	80	__all__ = ["get_cubin_dir", "list_cubins", "get_cubin_path", "CUBIN_DIR"]	CODE
LOW	tests/comm/test_nvshmem_allreduce.py	12	logger = logging.getLogger(__name__)	CODE
LOW	tests/comm/test_vllm_custom_allreduce.py	18	logger = logging.getLogger(__name__)	CODE
LOW	tests/gemm/test_cute_dsl_gemm_allreduce_two_shot.py	36	logger = logging.getLogger(__name__)	CODE
LOW	flashinfer-jit-cache/flashinfer_jit_cache/__init__.py	38	__all__ = [	CODE
LOW	examples/pytorch/wan/transformer_wan_flashinfer.py	60	logger = logging.getLogger(__name__)	CODE
LOW	scripts/aot_memory_monitor.py	265	def update_stats(stats: MemoryStats, sample: dict[str, int]) -> None:	CODE
LOW	flashinfer/fp8_quantization.py	20	__all__ = [	CODE
LOW	flashinfer/compilation_context.py	24	logger = logging.getLogger(__name__)	CODE
LOW	flashinfer/utils.py	736	def set_log_level(lvl_str: str) -> None:	CODE
LOW	flashinfer/fp4_quantization.py	43	__all__ = [	CODE
LOW	flashinfer/artifacts.py	289	def update_pbar_cb(_) -> None:	CODE
LOW	flashinfer/deep_gemm.py	45	logger = logging.getLogger(__name__)	CODE
LOW	flashinfer/attention/__init__.py	20	__all__ = [	CODE
LOW	flashinfer/trace/solution.py	278	__all__ = [	CODE
LOW	flashinfer/trace/__init__.py	42	__all__ = [	CODE
LOW	flashinfer/trace/templates/_init_helpers.py	258	__all__ = [	CODE
LOW	flashinfer/moe_ep/__init__.py	103	__all__ = [	CODE
LOW	flashinfer/moe_ep/layer.py	16	__all__ = ["MoEEpLayer", "MoEEpMegaLayer", "MoEEpSplitLayer"]	CODE
LOW	flashinfer/moe_ep/core/__init__.py	13	__all__ = [	CODE
LOW	flashinfer/moe_ep/core/bootstrap_utils.py	48	__all__ = [	CODE
LOW	flashinfer/moe_ep/core/runtime/bootstrap.py	258	__all__ = [	CODE
LOW	flashinfer/moe_ep/core/runtime/__init__.py	15	__all__ = [	CODE
LOW	flashinfer/moe_ep/core/comm/__init__.py	6	__all__ = ["Fleet", "Handle", "create_fleet"]	CODE
LOW	flashinfer/moe_ep/core/kernel/__init__.py	13	__all__ = [	CODE
LOW	flashinfer/moe_ep/core/validation/__init__.py	18	__all__ = [	CODE
LOW	flashinfer/moe_ep/modes/config.py	10	__all__ = [	CODE
LOW	flashinfer/moe_ep/modes/__init__.py	11	__all__ = [	CODE
LOW	flashinfer/moe_ep/backends/__init__.py	5	__all__ = ["mega", "split"]	CODE
LOW	flashinfer/moe_ep/backends/split/__init__.py	6	__all__ = [	CODE
LOW	flashinfer/moe_ep/backends/split/comm/__init__.py	8	__all__ = ["NCCLEPConfig", "NcclEpConfig", "NvepConfig"]	CODE
LOW	flashinfer/moe_ep/backends/split/comm/nccl_ep/fleet.py	39	logger = logging.getLogger(__name__)	CODE
LOW	flashinfer/moe_ep/backends/split/kernel/__init__.py	5	__all__ = ["fused_moe", "identity"]	CODE
LOW	…nfer/moe_ep/backends/split/kernel/identity/__init__.py	4	__all__ = ["IdentityConfig", "IdentitySplitKernelBackend"]	CODE
LOW	…fer/moe_ep/backends/split/kernel/fused_moe/__init__.py	4	__all__ = ["FusedMoeKernelConfig", "FusedMoeSplitKernelBackend"]	CODE
LOW	flashinfer/moe_ep/backends/mega/__init__.py	5	__all__ = ["kernel"]	CODE
LOW	flashinfer/moe_ep/backends/mega/kernel/__init__.py	5	__all__ = ["deep_gemm_mega", "mxfp8_cutedsl", "nvfp4_cutedsl"]	CODE
LOW	…infer/moe_ep/backends/mega/kernel/weight_validation.py	91	__all__ = ["check_transformed_mega_weights_structure", "check_transformed_weight_pair"]	CODE
LOW	…moe_ep/backends/mega/kernel/deep_gemm_mega/__init__.py	5	__all__ = [	CODE
LOW	…/moe_ep/backends/mega/kernel/deep_gemm_mega/weights.py	159	__all__ = [	CODE
LOW	…/moe_ep/backends/mega/kernel/deep_gemm_mega/staging.py	38	__all__ = ["stage_mega_moe_inputs"]	CODE
LOW	…/moe_ep/backends/mega/kernel/mxfp8_cutedsl/__init__.py	5	__all__ = [	CODE
LOW	…r/moe_ep/backends/mega/kernel/mxfp8_cutedsl/weights.py	350	__all__ = [	CODE
LOW	…ckends/mega/kernel/cutedsl_backend_kernels/__init__.py	11	__all__ = ["bootstrap_paths"]	CODE
LOW	…ga/kernel/cutedsl_backend_kernels/frontend/__init__.py	50	__all__ = [	CODE
LOW	…backend_kernels/frontend/megamoe_frontend/api_mxfp8.py	173	def set_gate_up_clamp(self, clamp: Optional[float]) -> None:	CODE
LOW	…_backend_kernels/frontend/megamoe_frontend/__init__.py	24	__all__ = [	CODE
LOW	…tedsl_backend_kernels/frontend/megamoe_frontend/api.py	23	__all__ = [	CODE
LOW	…backend_kernels/frontend/megamoe_frontend/api_nvfp4.py	231	def set_gate_up_clamp(self, clamp: Optional[float]) -> None:	CODE
LOW	…dsl_backend_kernels/moe_nvfp4_swapab/mega_reference.py	2613	__all__ = ["MegaMoEReference", "compute_megamoe_reference", "Nvfp4BlockSize"]	CODE
LOW⚡	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	869	def set_torch_profiler_enabled(self, enabled: bool) -> None:	CODE
LOW⚡	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	873	def set_perf_iters(self, warmup: int, iters: int) -> None:	CODE
LOW	…utedsl_backend_kernels/moe_nvfp4_swapab/kernel_fc12.py	270	def _setup_attributes(self) -> None:	CODE
LOW	…backend_kernels/moe_mxfp8_glu/kernel_mxfp8_glu_fc12.py	293	def _setup_attributes(self) -> None:	CODE
LOW	…_backend_kernels/moe_mxfp8_glu/mega_reference_mxfp8.py	180	__all__ = ["compute_megamoe_reference_mxfp8", "Mxfp8BlockSize"]	CODE
LOW	…/moe_ep/backends/mega/kernel/nvfp4_cutedsl/__init__.py	5	__all__ = [	CODE
LOW	…r/moe_ep/backends/mega/kernel/nvfp4_cutedsl/weights.py	314	__all__ = [	CODE
LOW	…ashinfer/gdn_kernels/gdn_decode_bf16_wy_output_only.py	2255	__all__ = ["gated_delta_rule_mtp", "GdnDecodeKernel", "K_DIM", "V_DIM_C"]	CODE
LOW	flashinfer/gdn_kernels/__init__.py	91	__all__ = [	CODE
LOW	flashinfer/gdn_kernels/blackwell/__init__.py	13	__all__ = [	CODE
77 more matches not shown…

Cross-Language Confusion23 hits · 102 pts

Severity	File	Line	Snippet	Context
HIGH	tests/comm/test_trtllm_allreduce_reduction.py	43	TORCH_CHECK(num_threads > 0 && num_threads <= 1024, "invalid block size");	CODE
HIGH	tests/comm/test_trtllm_mnnvl_allreduce.py	306	ordinary data; a `val == 0.F && signbit(val)` check would FTZ-flush	STRING
HIGH	tests/utils/test_jit_example.py	193	if (qo_idx < qo_len && kv_idx < kv_len) {	CODE
HIGH	benchmarks/bench_monomoe.py	86	Uses mono_moe's `[gate \|\| up]` fc1 half-ordering and the given (already	STRING
HIGH	benchmarks/routines/attention.py	2790	# Leave empty (null) when not explicitly overridden so legacy	COMMENT
HIGH	flashinfer/trace/templates/moe.py	2516	description="Up/gate projection weights (FP8, [gate \|\| up] rows).",	CODE
HIGH	flashinfer/trace/templates/moe.py	2367	up-projection weights use the [gate \|\| up] half ordering (monomoe's	STRING
HIGH	…tedsl_backend_kernels/frontend/megamoe_frontend/run.py	11	cd cutedsl_megamoe/front_end && pip install -e .	STRING
HIGH	…backend_kernels/moe_mxfp8_glu/kernel_mxfp8_glu_fc12.py	107	f"Expected one of: {[t.__name__ for t in valid_ab_tuple]}." # type: ignore[name-defined] # noqa: F	CODE
HIGH	…/mega/kernel/cutedsl_backend_kernels/src/flag_batch.py	21	flag_addr: Int64 # per-lane counter-slot address (0 == null)	CODE
HIGH⚡	…ashinfer/gdn_kernels/gdn_decode_bf16_wy_output_only.py	1028	# _v7e_a/b regs are undefined), so they must keep log_alpha=beta=0. The causal	COMMENT
HIGH	flashinfer/fused_moe/monomoe.py	23	N (== N_half) = 512 (up-projection produces 2*N = 1024 rows: gate \|\| up)	STRING
HIGH	flashinfer/fused_moe/cute_dsl/moe_utils.py	359	``!enable_alltoall \|\| ep_size <= top_k`` branch). Functionally	STRING
HIGH	flashinfer/mla/_core.py	3180	Accepts the concatenated ``(q_nope \|\| q_rope)`` query and ``(ckv \|\| kpe)`` paged KV	CODE
HIGH	flashinfer/jit/attention/variants.py	39	float log_sink = (kv_tile_idx == 0 && qo_head_idx < params.num_qo_heads) ? params.sink[qo_head_idx] * math::log2e :	CODE
HIGH	flashinfer/jit/attention/fmha_v2/utils.py	543	code = "if (!{has_noloop} \|\| (!force_unroll && (ignore_b1opt \|\| b > {unroll_threshold})))".format(	STRING
HIGH	flashinfer/jit/attention/fmha_v2/utils.py	547	code = "if (!{has_noloop} \|\| (!force_unroll && (ignore_b1opt \|\| b * h > {unroll_threshold})))".format(	STRING
HIGH	flashinfer/jit/attention/fmha_v2/generator_utils.py	1520	USE_TMA_STORE && false,	CODE
HIGH	flashinfer/jit/attention/fmha_v2/generator_utils.py	1541	USE_TMA_STORE && false,	CODE
HIGH	flashinfer/jit/attention/fmha_v2/generator_utils.py	2401	code = "if (!{has_noloop} \|\| (!force_unroll && (ignore_b1opt \|\| b > {unroll_threshold})))".format(	CODE
HIGH	flashinfer/jit/attention/fmha_v2/generator_utils.py	2405	code = "if (!{has_noloop} \|\| (!force_unroll && (ignore_b1opt \|\| b * h > {unroll_threshold})))".format(	CODE
HIGH	flashinfer/jit/attention/fmha_v2/fmha_library.py	761	code = "if (!{has_noloop} \|\| (!force_unroll && (ignore_b1opt \|\| b > {unroll_threshold})))".format(	CODE
HIGH	flashinfer/jit/attention/fmha_v2/fmha_library.py	765	code = "if (!{has_noloop} \|\| (!force_unroll && (ignore_b1opt \|\| b * h > {unroll_threshold})))".format(	CODE

Redundant / Tautological Comments58 hits · 87 pts

Severity	File	Line	Snippet	Context
LOW	build_backend.py	933	# Check if we're in a git repository	COMMENT
LOW	flashinfer-cubin/build_backend.py	75	# Check if we're in a git repository	COMMENT
LOW	tests/attention/test_attention_sink.py	918	# Check if any request has qo_len > kv_len for causal case	COMMENT
LOW	…s/comm/test_trtllm_allreduce_fusion_group_fp8_quant.py	218	# Check if current workspace is large enough; recreate if not	COMMENT
LOW	tests/utils/test_fp8_quantize.py	651	# Check if the bits of paddings are zero.	COMMENT
LOW	tests/utils/test_load_cubin_compile_race_condition.py	44	# Read the file from FLASHINFER_CUBIN_DIR	COMMENT
LOW	tests/model_optimizations/test_dsv3_fused_routing.py	274	# Check if kernel's selection matches expected experts (exact or tied)	COMMENT
LOW	tests/test_helpers/sink_attention_reference.py	95	# Check if variable length mode is indicated by presence of indptr	COMMENT
LOW	flashinfer-jit-cache/build_backend.py	57	# Check if we're in a git repository	COMMENT
LOW	benchmarks/bench_tgv_gemm.py	184	# Check if results are close enough	COMMENT
LOW	benchmarks/bench_logging_overhead.py	224	# Check if CUDA is available	COMMENT
LOW	benchmarks/bench_dcp_alltoall.py	297	# Print results (rank 0 only)	COMMENT
LOW	scripts/xfails_tracker.py	52	# Check if this is a test function (starts with "test_" or has @pytest marks)	COMMENT
LOW	scripts/xfails_tracker.py	70	# Check if this is a pytest.param call	COMMENT
LOW	scripts/xfails_tracker.py	77	# Check if this is a runtime pytest.xfail() call	COMMENT
LOW	scripts/test_utils.sh	1625	# Check if parallel execution is enabled	COMMENT
LOW	scripts/codeowner_analyzer.py	89	# Check if gh CLI is available	COMMENT
LOW	scripts/codeowner_analyzer.py	114	# Check if it's already in our cache	COMMENT
LOW	scripts/codeowner_analyzer.py	220	# Check if username is in allowed list (case-insensitive)	COMMENT
LOW	scripts/codeowner_analyzer.py	290	# Check if it's a relevant code file	COMMENT
LOW	scripts/codeowner_analyzer.py	420	# Check if contributor is in allowed users list	COMMENT
LOW	scripts/codeowner_analyzer.py	629	# Write file-level overrides LAST (CODEOWNERS uses last-match-wins)	COMMENT
LOW	scripts/task_run_unit_tests.sh	64	# Check if this file's directory should be excluded	COMMENT
LOW	.github/workflows/update-xfails-report.yml	43	# Check if xfails report file is new (unstaged) or has changes	COMMENT
LOW	.github/workflows/update-codeowners.yml	50	# Check if CODEOWNERS file is new (unstaged) or has changes	COMMENT
LOW	.github/workflows/pr-test.yml	80	# Check if PR has run-ci label	COMMENT
LOW	.github/workflows/pr-test.yml	87	# Check if PR author is a member of ci-users team	COMMENT
LOW	.github/workflows/issue-claim.yml	86	# Check if issue already has assignees	COMMENT
LOW	.github/workflows/issue-claim.yml	301	# Check if user is now a collaborator (accepted invitation)	COMMENT
LOW	flashinfer/api_logging.py	2083	# Check if this parameter was provided	COMMENT
LOW	flashinfer/utils.py	300	# Check if it's a CUDA device	COMMENT
LOW	flashinfer/prefill.py	523	# Check if FP8 by presence of scale tensors	COMMENT
LOW	flashinfer/__main__.py	460	# Check if this is a single dump or a session / sequence root	COMMENT
LOW	flashinfer/cuda_utils.py	20	# Check if cuda.cudart module is available and import accordingly	COMMENT
LOW	flashinfer/autotuner/autotuner.py	1582	# Set time_measured to inf to notify the failure of the tactic. This can happen when	COMMENT
LOW	flashinfer/triton/sm_constraint_gemm.py	66	# Set num_sms to be 100% of the available SMs	COMMENT
LOW	flashinfer/comm/mnnvl.py	988	# Check if cuda.cudart is available and import accordingly	COMMENT
LOW	flashinfer/comm/mnnvl.py	1025	# Check if device supports multicasting	COMMENT
LOW	…ockscaled_contiguous_gather_grouped_gemm_act_fusion.py	510	# Check if we're doing FP4 quantization	COMMENT
LOW	…hinfer/fused_moe/cute_dsl/blackwell/custom_pipeline.py	357	# Set mask to None if not using clusters (i.e. 1CTA kernels)	COMMENT
LOW	…hinfer/fused_moe/cute_dsl/blackwell/custom_pipeline.py	363	# Set mask to None if not using 2CTA instructions	COMMENT
LOW	…blockscaled_contiguous_grouped_gemm_finalize_fusion.py	688	# Check if input data types are compatible with MMA instruction	COMMENT
LOW	…ockscaled_contiguous_gather_grouped_gemm_act_fusion.py	853	# Check if input data types are compatible with MMA instruction	COMMENT
LOW	flashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py	359	# Check if input data types are compatible with MMA instruction	COMMENT
LOW	flashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py	994	# Check if we're not at the last buffer of the last k_tile	COMMENT
LOW	flashinfer/gemm/kernels/dense_blockscaled_gemm_sm103.py	1079	# Check if we're not at the last stage of the last k_tile	COMMENT
LOW	…ashinfer/gemm/kernels/grouped_gemm_masked_blackwell.py	904	# Check if input data types are compatible with MMA instruction	COMMENT
LOW	…er/gemm/kernels/dense_blockscaled_gemm_sm100_common.py	102	# Check if input data types are compatible with MMA instruction	COMMENT
LOW	flashinfer/quantization/kernels/mxfp8_quantize.py	605	# Check if this thread's row is valid	COMMENT
LOW	flashinfer/cute_dsl/gemm_allreduce_two_shot.py	519	# Check if input data types are compatible with MMA instruction	COMMENT
LOW	flashinfer/testing/utils.py	867	# Check if args are provided (determines how we call fn)	COMMENT
LOW	flashinfer/testing/utils.py	1164	# Check if args are provided (determines how we call fn)	COMMENT
LOW	flashinfer/testing/utils.py	1432	# Check if args are provided (determines how we call fn)	COMMENT
LOW	flashinfer/jit/cpp_ext.py	213	# Check if module provides architecture flags	COMMENT
LOW	flashinfer/jit/fused_moe.py	253	# Check if checksums.txt exists in the cubin directory	COMMENT
LOW	flashinfer/jit/attention/modules.py	1869	# Check if checksums.txt exists in the cubin directory	COMMENT
LOW	flashinfer/jit/gemm/core.py	673	# Check if checksums.txt exists in the cubin directory	COMMENT
LOW	flashinfer/jit/gemm/core.py	847	# Check if checksums.txt exists in the cubin directory	COMMENT

Docstring Block Structure13 hits · 65 pts

Severity	File	Line	Snippet	Context
HIGH	flashinfer/green_ctx.py	129	Split the device into multiple `green contexts <https://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__GREEN__CO	STRING
HIGH	flashinfer/green_ctx.py	199	Split the device into multiple green contexts, each with a fixed number of SMs, return the corresponding stream	STRING
HIGH	flashinfer/utils.py	284	Get GPU memory bandwidth in GB/s for the specified CUDA device. Args: device: torch.device object, e.g	STRING
HIGH	flashinfer/autotuner/autotuner.py	96	Map x to the nearest bucket using floor or ceil semantics. Args: x: The value to map. buckets: Bu	STRING
HIGH	flashinfer/autotuner/autotuner.py	140	Create a mapper function for :class:`DynamicTensorSpec.map_to_tuning_buckets`. The returned callable maps any integ	STRING
HIGH	flashinfer/autotuner/autotuner.py	578	Context manager for autotuning with optional file-based caching. Controls how FlashInfer profiles and selects the b	STRING
HIGH	flashinfer/autotuner/autotuner.py	2092	Load autotuner configs from a JSON file. Populates the internal config lookup table so that ``search_cache()``	STRING
HIGH	…blockscaled_contiguous_grouped_gemm_finalize_fusion.py	283	Blockscaled Contiguous Grouped GEMM with Finalize Fusion for MoE workloads. Performs grouped matrix multiplication	STRING
HIGH	…ockscaled_contiguous_gather_grouped_gemm_act_fusion.py	362	Blockscaled Contiguous Gather Grouped GEMM with SwiGLU Fusion for MoE workloads. Performs grouped matrix multiplica	STRING
HIGH	flashinfer/gemm/gemm_bf16_fp4.py	130	Prepare FP4 weights for the bf16 x fp4 GEMM, for a specific backend. The caller is expected to start with weights i	STRING
HIGH	flashinfer/quantization/fp4_quantization.py	408	Quantize a batched tensor to FP4 (E2M1x2) with per-block scale factors. This function converts a float/bfloat16	STRING
HIGH	flashinfer/quantization/fp4_quantization.py	556	Quantize a silu and matmul with masked batched tensor to FP4 (E2M1x2) with per-block scale factors. This functi	STRING
HIGH	flashinfer/parallel_attention/utils.py	356	Create a device mesh and return the Ring and Ulysses process groups. Builds a ``DeviceMesh`` with up to three dimen	STRING

AI Slop Vocabulary23 hits · 60 pts

Severity	File	Line	Snippet	Context
LOW⚡	tests/trace/test_fi_trace_template_consistency.py	313	# To add a new kernel: no changes needed here — simply add	COMMENT
MEDIUM	tests/autotuner/test_autotuner_distributed.py	220	# independent of how the harness imported the tests (a ``spawn`` worker that	COMMENT
MEDIUM	tests/mamba/test_philox_rounding.py	785	"""Software e4m3 SR matches hardware PTX path bitwise across a comprehensive	STRING
MEDIUM⚡	tests/gdn/test_decode_delta_rule.py	1266	# Test MTP kernel with FP32 state, cache ON, state update ON (comprehensive)	COMMENT
LOW	tests/test_helpers/alibi_reference.py	78	# Since it's causal mask we can just use $[0, 1, \dots, N]$ too.	STRING
MEDIUM⚡	tests/moe/test_unified_moe_fuzz.py	692	# different shape would produce a wrong answer here. (Shares the harness's snap/reference/prep.)	COMMENT
MEDIUM⚡	benchmarks/bench_mamba_ssd_combined.py	257	# Layer 2: bench_mode — batched or varlen harness	COMMENT
MEDIUM	scripts/task_test_moe_ep_smoke.sh	4	# Run the moe_ep smoke + multi-rank harness via torchrun.	COMMENT
LOW	flashinfer/api_logging.py	1547	# The safest way is to just call the class constructor.	COMMENT
MEDIUM	flashinfer/aot.py	277	# The holistic (persistent) batch-attention kernel	COMMENT
MEDIUM	flashinfer/attention/_core.py	160	# head_dim > 256 is for holistic (persistent) kernel.	STRING
MEDIUM⚡	…_backend_kernels/moe_nvfp4_swapab/epilogue_refactor.py	517	_EpilogueTokenTileSize = 64 # Fundamentally the epi_tile_n	CODE
MEDIUM⚡	…_backend_kernels/moe_nvfp4_swapab/epilogue_refactor.py	518	_EpilogueFc1IntermediateGateUpTileSize = 128 # Fundamentally epi_tile_m	CODE
MEDIUM⚡	…_backend_kernels/moe_nvfp4_swapab/epilogue_refactor.py	519	_EpilogueFc1IntermediateDownTileSize = 64 # Fundamentally epi_tile_m // 2	CODE
MEDIUM⚡	…_backend_kernels/moe_nvfp4_swapab/epilogue_refactor.py	520	_EpilogueFc2HiddenTileSize = 128 # Fundamentally epi_tile_m	CODE
MEDIUM	…ckend_kernels/moe_nvfp4_swapab/run_functional_tests.sh	2	# Functional test harness for the local fused fc1+fc2 MoE NVFP4 swap-AB runner.	COMMENT
MEDIUM	…dsl_backend_kernels/moe_nvfp4_swapab/run_mega_tests.sh	2	# Functional test harness for the distributed MegaMoE fused dispatch + fc1/fc2	COMMENT
MEDIUM	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	2486	# to avoid recomputing it the harness keeps the global	COMMENT
MEDIUM	…_backend_kernels/moe_mxfp8_glu/run_functional_tests.sh	2	# Functional test harness for the MXFP8 GLU fused fc1+fc2 runner.	COMMENT
MEDIUM	…utedsl_backend_kernels/moe_mxfp8_glu/run_mega_tests.sh	2	# Functional test harness for the MXFP8 distributed MegaMoE fused dispatch +	COMMENT
MEDIUM	flashinfer/gemm/gemm_base.py	6211	# stays row-major (m, n). Pass swap_ab=False to the harness so it keeps	STRING
MEDIUM	flashinfer/jit/monomoe.py	123	# robust choice for a kernel that is SM90a-only by construction (it	COMMENT
MEDIUM	flashinfer/jit/attention/fmha_v2/generator_utils.py	1760	// distribute m steps to multiple blocks (fully utilize SMs)	COMMENT

AI Response Leakage3 hits · 22 pts

Severity	File	Line	Snippet	Context
HIGH	docs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py	304	# In this example, you should learn how minimal the JAX-TVM bridge can be when the function signature and ``arg_spec`` a	COMMENT
HIGH	docs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py	427	# In this example, you should learn how to handle two new complications that were absent in Example 1: a TVM function th	COMMENT
HIGH	docs/tutorials/jax_tvm_ffi/flashinfer_jax_tvm_ffi.py	644	# In this example, you should learn about three patterns that appear together for the first time: a kernel **compiled se	COMMENT

Modern AI Meta-Vocabulary6 hits · 21 pts

Severity	File	Line	Snippet	Context
MEDIUM⚡	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	2893	# Top-level orchestration	COMMENT
MEDIUM	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	1151	# P0 scaffolding: tile-wise low-precision combine staging buffers.	COMMENT
MEDIUM	…utedsl_backend_kernels/moe_nvfp4_swapab/mega_runner.py	3167	# orchestration above is the part being smoke-tested for now.	COMMENT
MEDIUM⚡	flashinfer/cute_dsl/attention/roles/epilogue.py	80	# Prefill orchestration — proven-correct inline implementation	COMMENT
MEDIUM⚡	flashinfer/cute_dsl/attention/roles/loader_tma.py	141	# Prefill orchestration — proven-correct inline implementation	COMMENT
MEDIUM⚡	flashinfer/cute_dsl/attention/roles/mma.py	202	# Prefill orchestration — uses primitives for GEMMs and TMEM lifecycle	COMMENT

Overly Generic Function Names13 hits · 12 pts

Severity	File	Line	Snippet	Context
LOW	tests/utils/test_decorators.py	15	def my_function(x, y):	CODE
LOW	tests/utils/test_logging.py	109	def test_function(x, y):	CODE
LOW	tests/utils/test_logging.py	135	def test_function(tensor, value):	CODE
LOW	tests/utils/test_logging.py	173	def test_function(tensor):	CODE
LOW	tests/utils/test_logging.py	201	def test_function(mode: TestEnum, strategy: StringEnum):	CODE
LOW	tests/utils/test_logging.py	231	def test_function(x, y=10, z=20, mode=TestEnum.OPTION_A):	CODE
LOW	tests/utils/test_logging.py	262	def test_function(x, y=10, z=20):	CODE
LOW	tests/utils/test_logging.py	359	def test_function(	CODE
LOW	tests/utils/test_logging.py	397	def test_function(tensor):	CODE
LOW	tests/utils/test_logging.py	430	def test_function(nested):	CODE
LOW	tests/utils/test_logging.py	495	def test_function(x):	CODE
LOW	tests/utils/test_logging.py	520	def test_function(a, b, c):	CODE
LOW	flashinfer/api_logging.py	2415	... def my_function(x, y):	STRING

Example Usage Blocks6 hits · 9 pts

Severity	File	Line	Snippet	Context
LOW	csrc/fmha_v2/fmha/utils.h	66	// Example usage:	COMMENT
LOW	scripts/list_apis.sh	6	# Usage:	COMMENT
LOW	…ckend_kernels/moe_nvfp4_swapab/run_functional_tests.sh	10	# Usage:	COMMENT
LOW	…dsl_backend_kernels/moe_nvfp4_swapab/run_mega_tests.sh	17	# Usage:	COMMENT
LOW	…_backend_kernels/moe_mxfp8_glu/run_functional_tests.sh	18	# Usage:	COMMENT
LOW	…utedsl_backend_kernels/moe_mxfp8_glu/run_mega_tests.sh	22	# Usage:	COMMENT

Synthetic Comment Markers1 hit · 5 pts

Severity	File	Line	Snippet	Context
HIGH	.github/workflows/update-codeowners.yml	76	Auto-generated CODEOWNERS update based on commit activity over the last ${{ env.DAYS_BACK }} days.	CODE

Fake / Example Data2 hits · 2 pts

Severity	File	Line	Snippet	Context
LOW	flashinfer/comm/mnnvl.py	798	dummy_data = b"\x00"	CODE
LOW	flashinfer/comm/mnnvl.py	805	self.sock.sendmsg([dummy_data], ancillary, 0, dest_path)	CODE

Analysis Overview

What These Metrics Mean

Score History

Severity Breakdown

Directory Score Breakdown

Pattern Findings