pytorch/torchtitan

20.0

Adjusted Score

20.0

Raw Score

100%

Time Factor

2026-08-29

Last Push

5.7K

Stars

Python

Language

176.0K

Lines of Code

665

Files

3.0K

Pattern Hits

2026-08-29

Scan Date

0.04

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 28MEDIUM 210LOW 2747

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 2985 distinct pattern matches across 20 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.

Hyper-Verbose Identifiers1792 hits · 1674 pts

Severity	File	Line	Snippet	Context
LOW	tests/unit_tests/test_kda_attention.py	87	def test_varlen_matches_independent_documents(self):	CODE
LOW	tests/unit_tests/test_qwen3_5_mrope_positions.py	104	def test_multimodal_batch_routes_mrope_to_layers(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu.py	76	def test_gate_up_projection_is_linear(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu.py	85	def test_preserves_converted_linear_config(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu.py	97	def test_saves_in_stock_layout(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu.py	106	def test_fused_checkpoint_loads_into_stock(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu.py	118	def test_stock_checkpoint_loads_into_fused(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu.py	148	def test_strict_load_reports_missing(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu.py	172	def test_dist_gemm_config_keeps_overlap(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu.py	202	def test_hf_adapter_roundtrip(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_compile_regional_inductor.py	36	def test_inductor_backend_left_unchanged(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_compile_regional_inductor.py	41	def test_aot_eager_with_flex_scoops(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_compile_regional_inductor.py	46	def test_aot_eager_without_flex_left_unchanged(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_compile_regional_inductor.py	51	def test_other_backend_with_flex_raises(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_compile_regional_inductor.py	56	def test_region_annotation_follows_flag(self):	CODE
LOW	tests/unit_tests/gpu/test_compile_regional_inductor.py	90	def _triton_flex_kernel_count(backend) -> int:	CODE
LOW	tests/unit_tests/gpu/test_compile_regional_inductor.py	118	def test_plain_aot_eager_decomposes_flex(self):	CODE
LOW	tests/unit_tests/gpu/test_compile_regional_inductor.py	122	def test_regional_scoop_lowers_flex_to_triton(self):	CODE
LOW	tests/unit_tests/gpu/test_activation_checkpoint.py	218	def test_force_recompute_mm_fqns(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu_override.py	31	def _build_fused_grouped_experts() -> FusedGroupedExperts:	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu_override.py	44	def test_minimal_async_ep_config_imports_override(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu_override.py	52	def test_grouped_experts_config_is_replaced(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu_override.py	68	def test_saves_and_loads_stock_layout(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu_override.py	82	def test_built_module_has_only_fused_params(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu_override.py	143	def test_matches_unfused_forward_and_backward(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu_override.py	195	def test_silu_and_mul_custom_op_matches_reference_with_offsets(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu_override.py	216	def test_silu_and_mul_custom_op_matches_reference_without_offsets(self):	CODE
LOW	tests/unit_tests/gpu/test_fused_swiglu_override.py	236	def test_silu_and_mul_kernels_match_reference_with_offsets(self):	CODE
LOW	tests/unit_tests/gpu/test_kimi_k3.py	121	def test_attention_gym_kda_kernel_matches_recurrent_reference(self):	CODE
LOW	tests/unit_tests/gpu/test_kimi_k3.py	193	def test_state_dict_round_trips_through_hf_adapter(self):	CODE
LOW	tests/unit_tests/gpu/test_minimal_async_ep_kernels.py	36	def test_topk_index_kernels_match_reference(self):	CODE
LOW	tests/unit_tests/gpu/test_minimal_async_ep_kernels.py	192	def test_metadata_kernels_match_reference(self):	CODE
LOW	tests/unit_tests/gpu/test_minimal_async_ep_kernels.py	298	def test_copy_kernels_handle_strides_and_active_row_masks(self):	CODE
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	27	def _torch_native_gated_delta(	CODE
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	76	def _torch_native_gated_delta_varlen(	CODE
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	171	def test_flex_masks_ignore_padding_position_resets(self):	CODE
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	189	def test_flex_masks_include_delta_net_varlen_metadata(self):	CODE
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	377	def test_extracted_forward_matches_main(self):	CODE
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	399	def _assert_packed_run_matches_per_document(self, model, x, positions, masks):	CODE
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	420	def test_varlen_matches_independent_document_forwards(self):	CODE
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	437	def test_get_attention_masks_pairs_flex_mask_with_deltanet_offsets(self):	CODE
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	506	def _assert_fla_varlen_matches_per_document(	CODE
LOW⚡	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	590	def test_fla_chunked_varlen_matches_independent_document_forwards(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	598	def test_fla_fused_recurrent_varlen_matches_independent_document_forwards(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	603	def test_varlen_offsets_are_fresh_per_deltanet_invocation(self):	CODE
LOW	tests/unit_tests/gpu/test_qk_clip.py	38	def test_attention_records_training_maxima_only(self) -> None:	CODE
LOW	tests/unit_tests/gpu/test_qk_clip.py	86	def test_optimizer_hook_runs_qk_clip(self) -> None:	CODE
LOW	tests/unit_tests/gpu/test_qk_clip.py	137	def test_reduces_head_maxima_and_clips_local_dtensor_weights(self) -> None:	CODE
LOW	tests/unit_tests/gpu/test_qk_clip.py	230	def test_weight_scaling_is_communication_free(self) -> None:	CODE
LOW	tests/unit_tests/gpu/test_compile_moe.py	46	def test_async_tp_requires_model_compile(self):	CODE
LOW	tests/unit_tests/gpu/test_compile_moe.py	64	def test_apply_compile_configures_async_tp(self):	CODE
LOW	tests/unit_tests/gpu/test_compile_moe.py	94	def test_grouped_mm_compiles_and_runs(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_helion_rope.py	86	def test_registered_against_cossin(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_helion_rope.py	91	def test_registered_against_complex(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_helion_rope.py	96	def test_cossin_factory_preserves_fields(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_helion_rope.py	296	def test_forward_matches_cossin(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_helion_rope.py	305	def test_forward_matches_with_implicit_positions(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_helion_rope.py	313	def test_complex_forward_matches(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_helion_rope.py	322	def test_complex_forward_matches_with_implicit_positions(self):	CODE
LOW⚡	tests/unit_tests/gpu/test_helion_rope.py	329	def test_backward_matches_cossin(self):	CODE
1732 more matches not shown…

Decorative Section Separators165 hits · 571 pts

Severity	File	Line	Snippet	Context
MEDIUM⚡	tests/unit_tests/cpu/test_tp_kv_heads_validation.py	109	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/unit_tests/cpu/test_tp_kv_heads_validation.py	111	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/unit_tests/cpu/test_tp_kv_heads_validation.py	119	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/unit_tests/cpu/test_tp_kv_heads_validation.py	121	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/unit_tests/cpu/test_tp_kv_heads_validation.py	129	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	tests/unit_tests/cpu/test_tp_kv_heads_validation.py	131	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	86	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	88	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	319	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	321	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	510	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	512	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	649	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	651	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	705	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	707	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	755	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	757	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	1525	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	…nit_tests/cpu/observability/test_structured_logging.py	1527	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	47	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	49	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	298	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	300	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	544	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	546	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	790	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	792	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	990	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	992	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	1209	# ------------------------------------------------------------------	COMMENT
MEDIUM	…nit_tests/cpu/observability/test_structured_logging.py	1215	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	scripts/loss_compare.py	332	# =============================================================================	COMMENT
MEDIUM⚡	scripts/loss_compare.py	334	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	78	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	80	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	105	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	107	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	241	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	243	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	445	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	447	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	530	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	532	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	605	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	607	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	928	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	930	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	972	# =============================================================================	COMMENT
MEDIUM	scripts/loss_compare.py	974	# =============================================================================	COMMENT
MEDIUM⚡	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	47	# ============================================================	COMMENT
MEDIUM⚡	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	49	# ============================================================	COMMENT
MEDIUM	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	76	# ============================================================	COMMENT
MEDIUM	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	78	# ============================================================	COMMENT
MEDIUM	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	230	# ============================================================	COMMENT
MEDIUM	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	232	# ============================================================	COMMENT
MEDIUM	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	263	# ============================================================	COMMENT
MEDIUM	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	265	# ============================================================	COMMENT
MEDIUM	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	322	# ============================================================	COMMENT
MEDIUM	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	324	# ============================================================	COMMENT
105 more matches not shown…

Unused Imports321 hits · 308 pts

Severity	File	Line	Context
LOW	torchtitan/overrides/helion_rope.py	37	CODE
LOW	torchtitan/overrides/moe_token_dispatcher.py	40	CODE
LOW	torchtitan/tools/filesystem.py	15	CODE
LOW	torchtitan/distributed/spmd_types.py	9	CODE
LOW	torchtitan/distributed/fsdp.py	25	CODE
LOW	torchtitan/distributed/linear.py	28	CODE
LOW	torchtitan/distributed/__init__.py	7	CODE
LOW	torchtitan/distributed/utils.py	7	CODE
LOW	torchtitan/distributed/parallel_dims.py	7	CODE
LOW	…n/distributed/flex_shard/_optimizer_reshard_runtime.py	16	CODE
LOW	torchtitan/distributed/flex_shard/dist_muon.py	9	CODE
LOW	torchtitan/distributed/flex_shard/__init__.py	9	CODE
LOW	torchtitan/distributed/flex_shard/__init__.py	10	CODE
LOW	torchtitan/distributed/flex_shard/__init__.py	10	CODE
LOW	torchtitan/distributed/flex_shard/__init__.py	10	CODE
LOW	torchtitan/distributed/flex_shard/__init__.py	10	CODE
LOW	…/distributed/flex_shard/_optimizer_reshard_schedule.py	9	CODE
LOW	torchtitan/distributed/flex_shard/optimizer_reshard.py	9	CODE
LOW	torchtitan/distributed/context_parallel/__init__.py	19	CODE
LOW	torchtitan/distributed/context_parallel/__init__.py	19	CODE
LOW	torchtitan/distributed/context_parallel/__init__.py	19	CODE
LOW	torchtitan/distributed/context_parallel/api.py	25	CODE
LOW	torchtitan/distributed/minimal_async_ep/__init__.py	7	CODE
LOW	torchtitan/distributed/minimal_async_ep/__init__.py	7	CODE
LOW	torchtitan/distributed/minimal_async_ep/__init__.py	7	CODE
LOW	torchtitan/distributed/minimal_async_ep/__init__.py	7	CODE
LOW	torchtitan/distributed/minimal_async_ep/__init__.py	7	CODE
LOW	torchtitan/config/__init__.py	15	CODE
LOW	torchtitan/config/__init__.py	15	CODE
LOW	torchtitan/config/__init__.py	15	CODE
LOW	torchtitan/config/__init__.py	15	CODE
LOW	torchtitan/config/__init__.py	15	CODE
LOW	torchtitan/config/__init__.py	22	CODE
LOW	torchtitan/config/__init__.py	23	CODE
LOW	torchtitan/config/__init__.py	24	CODE
LOW	torchtitan/config/__init__.py	25	CODE
LOW	torchtitan/config/__init__.py	25	CODE
LOW	torchtitan/config/__init__.py	25	CODE
LOW	torchtitan/config/__init__.py	25	CODE
LOW	torchtitan/config/__init__.py	25	CODE
LOW	torchtitan/config/__init__.py	25	CODE
LOW	torchtitan/config/override.py	32	CODE
LOW	…formers_modeling_backend/tests/test_moe_parallelism.py	60	CODE
LOW	torchtitan/experiments/graph_trainer/precompile.py	7	CODE
LOW	torchtitan/experiments/graph_trainer/precompile.py	101	CODE
LOW	torchtitan/experiments/graph_trainer/fsdp_passes.py	14	CODE
LOW	torchtitan/experiments/graph_trainer/ep_eager_chunk.py	14	CODE
LOW	torchtitan/experiments/graph_trainer/custom_codegen.py	9	CODE
LOW	torchtitan/experiments/graph_trainer/ep_pass_utils.py	21	CODE
LOW	torchtitan/experiments/graph_trainer/registry.py	14	CODE
LOW	torchtitan/experiments/graph_trainer/graph_utils.py	11	CODE
LOW	torchtitan/experiments/graph_trainer/memory_policy.py	16	CODE
LOW	torchtitan/experiments/graph_trainer/memory_policy.py	52	CODE
LOW	torchtitan/experiments/graph_trainer/inductor_passes.py	15	CODE
LOW	torchtitan/experiments/graph_trainer/storage.py	7	CODE
LOW	…tan/experiments/graph_trainer/ep_process_group_pass.py	17	CODE
LOW	…eriments/graph_trainer/log_activation_memory_policy.py	9	CODE
LOW	torchtitan/experiments/graph_trainer/ep_chunk_pass.py	62	CODE
LOW	torchtitan/experiments/graph_trainer/passes.py	26	CODE
LOW	torchtitan/experiments/graph_trainer/passes.py	80	CODE
261 more matches not shown…

Structural Annotation Overuse97 hits · 139 pts

Severity	File	Line	Snippet	Context
LOW	.ci/docker/common/install_conda.sh	55	# WARNING: This is a HACK from PyTorch core to be able to build PyTorch on 22.04.	COMMENT
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	366	# Step 2: Load tokenizer using our function	COMMENT
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	372	# Step 3: Load tokenizer using official Tokenizer library (if available)	COMMENT
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	379	# Step 4: Load tokenizer using transformers AutoTokenizer (if available)	COMMENT
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	388	# Step 5: Compare underlying tokenizer attributes (only if official tokenizer is available)	COMMENT
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	394	# Step 6: Compare with transformers tokenizer if available	COMMENT
LOW	tests/unit_tests/cpu/test_tokenizer.py	351	# Step 1: Download tokenizer files	COMMENT
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	103	0.5, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	104	1.0, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	105	1.0, # Step 2: We maunally added step of stable phase, to prevent LR from dropping to 0 at last step	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	106	7.0 / 8.0, # Step 3: 7/8 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	107	6.0 / 8.0, # Step 4: 3/4 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	108	5.0 / 8.0, # Step 5: 5/8 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	109	4.0 / 8.0, # Step 6: 1/2 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	110	3.0 / 8.0, # Step 7: 3/8 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	111	2.0 / 8.0, # Step 8: 1/4 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	112	1.0 / 8.0, # Step 9: 1/8 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	146	0.5, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	147	1.0, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	148	1.0, # Step 2: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	149	1.0, # Step 3: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	150	1.0, # Step 4: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	151	1.0, # Step 5: We maunally added step of stable phase, to prevent LR from dropping to 0 at last step	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	152	0.8, # Step 6: Linear decay starts (80% of max LR)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	153	0.6, # Step 7: 60% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	154	0.4, # Step 8: 40% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	155	0.2, # Step 9: 20% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	212	0.2, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	213	0.4, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	214	0.6, # Step 2: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	215	0.8, # Step 3: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	216	1.0, # Step 4: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	249	0.5, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	250	1.0, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	251	1.0, # Step 2: We maunally added step of stable phase, to prevent LR from dropping to 0 at last step	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	252	1.0, # Step 3: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	253	1.0, # Step 4: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	254	1.0, # Step 5: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	255	1.0, # Step 6: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	256	1.0, # Step 7: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	257	1.0, # Step 8: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	258	1.0, # Step 9: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	292	0.2, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	293	0.4, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	294	0.6, # Step 2: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	295	0.8, # Step 3: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	296	1.0, # Step 4: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	297	1.0, # Step 5: We maunally added step of stable phase, to prevent LR from dropping to 0 at last step	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	298	0.8, # Step 6: Linear decay starts (80% of max LR)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	299	0.6, # Step 7: 60% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	300	0.4, # Step 8: 40% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	301	0.2, # Step 9: 20% of max LR	CODE
LOW⚡	scripts/checkpoint_compat_test.py	228	# Step 1: Reference run at load_commit (full training from scratch)	COMMENT
LOW⚡	scripts/checkpoint_compat_test.py	237	# Step 2: Save run at save_commit (partial training + checkpoint)	COMMENT
LOW	scripts/checkpoint_compat_test.py	253	# Step 3: Resume run at load_commit (load checkpoint, train to end)	COMMENT
LOW	scripts/checkpoint_compat_test.py	268	# Step 4: Compare	COMMENT
LOW⚡	.github/workflows/integration_test_8gpu_torchft.yaml	32	# Step 1: Dynamically compute the matrix based on conditions	COMMENT
LOW⚡	.github/workflows/integration_test_8gpu_torchft.yaml	40	# Step 2: Use the dynamic matrix in the build-test job	COMMENT
LOW⚡	.github/workflows/integration_test_8gpu_rl.yaml	30	# Step 1: Dynamically compute the matrix based on conditions	COMMENT
LOW⚡	.github/workflows/integration_test_8gpu_rl.yaml	40	# Step 2: Use the dynamic matrix in the build-test job	COMMENT
37 more matches not shown…

Deep Nesting147 hits · 136 pts

Severity	File	Line	Context
LOW	tests/unit_tests/gpu/test_fsdp_moe_sharding.py	67	CODE
LOW	tests/unit_tests/cpu/test_tokenizer.py	118	CODE
LOW	tests/unit_tests/cpu/test_tokenizer.py	474	CODE
LOW	tests/unit_tests/cpu/test_tokenizer.py	496	CODE
LOW	tests/unit_tests/cpu/test_tokenizer.py	519	CODE
LOW	tests/unit_tests/cpu/test_dataset_checkpointing.py	32	CODE
LOW	tests/unit_tests/cpu/test_embedding.py	110	CODE
LOW	tests/unit_tests/cpu/test_validator_frequency.py	13	CODE
LOW	tests/unit_tests/cpu/test_loss.py	360	CODE
LOW	tests/integration_tests/run_tests.py	137	CODE
LOW	tests/integration_tests/run_tests.py	390	CODE
LOW	tests/assets/cc12m_test/pack_test_dataset.py	13	CODE
LOW	…h_trace_compaction/scripts/compact_cuda_graph_trace.py	690	CODE
LOW	…h_trace_compaction/scripts/compact_cuda_graph_trace.py	740	CODE
LOW	…skills/numerics_debugging/scripts/activation_tracer.py	424	CODE
LOW	…skills/numerics_debugging/scripts/activation_tracer.py	455	CODE
LOW	…/skills/numerics_debugging/scripts/compare_numerics.py	82	CODE
LOW	…/skills/numerics_debugging/scripts/compare_numerics.py	341	CODE
LOW	…/skills/numerics_debugging/scripts/compare_numerics.py	489	CODE
LOW	…/skills/numerics_debugging/scripts/compare_numerics.py	1084	CODE
LOW	…/skills/numerics_debugging/scripts/compare_numerics.py	573	CODE
LOW	scripts/loss_compare.py	610	CODE
LOW	scripts/download_hf_assets.py	12	CODE
LOW	.github/scripts/todo_debt_tracker.py	120	CODE
LOW	.github/scripts/todo_debt_tracker.py	237	CODE
LOW	.github/scripts/todo_debt_tracker.py	338	CODE
LOW	torchtitan/trainer.py	274	CODE
LOW	torchtitan/trainer.py	774	CODE
LOW	torchtitan/trainer.py	943	CODE
LOW	torchtitan/trainer.py	198	CODE
LOW	torchtitan/trainer.py	219	CODE
LOW	torchtitan/tools/utils.py	140	CODE
LOW	torchtitan/distributed/spmd_types.py	231	CODE
LOW	torchtitan/distributed/fsdp.py	168	CODE
LOW	torchtitan/distributed/activation_checkpoint.py	31	CODE
LOW	torchtitan/distributed/activation_checkpoint.py	72	CODE
LOW	torchtitan/distributed/pipeline_parallel.py	357	CODE
LOW	torchtitan/distributed/pipeline_parallel.py	470	CODE
LOW	…n/distributed/flex_shard/_optimizer_reshard_runtime.py	128	CODE
LOW	…n/distributed/flex_shard/_optimizer_reshard_runtime.py	419	CODE
LOW	torchtitan/distributed/flex_shard/dist_muon.py	1134	CODE
LOW	torchtitan/distributed/flex_shard/dist_muon.py	1397	CODE
LOW	torchtitan/distributed/flex_shard/dist_muon.py	486	CODE
LOW	…/distributed/flex_shard/_optimizer_reshard_schedule.py	812	CODE
LOW	…/distributed/flex_shard/_optimizer_reshard_schedule.py	943	CODE
LOW	torchtitan/config/configurable.py	162	CODE
LOW	torchtitan/config/configurable.py	46	CODE
LOW	torchtitan/config/configurable.py	75	CODE
LOW	torchtitan/config/configurable.py	49	CODE
LOW	torchtitan/config/manager.py	48	CODE
LOW	…nts/transformers_modeling_backend/module_conversion.py	42	CODE
LOW	…tan/experiments/transformers_modeling_backend/model.py	815	CODE
LOW	…tan/experiments/transformers_modeling_backend/model.py	524	CODE
LOW	…tan/experiments/transformers_modeling_backend/model.py	894	CODE
LOW	…/experiments/transformers_modeling_backend/pipeline.py	39	CODE
LOW	…/experiments/transformers_modeling_backend/pipeline.py	148	CODE
LOW	…/experiments/transformers_modeling_backend/pipeline.py	191	CODE
LOW	…ments/transformers_modeling_backend/moe_replacement.py	92	CODE
LOW	…_moe_model/scripts/numerical_equivalence_pretrained.py	147	CODE
LOW	…/skills/add_moe_model/scripts/numerical_equivalence.py	523	CODE
87 more matches not shown…

Verbosity Indicators87 hits · 122 pts

Severity	File	Line	Snippet	Context
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	366	# Step 2: Load tokenizer using our function	COMMENT
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	372	# Step 3: Load tokenizer using official Tokenizer library (if available)	COMMENT
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	379	# Step 4: Load tokenizer using transformers AutoTokenizer (if available)	COMMENT
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	388	# Step 5: Compare underlying tokenizer attributes (only if official tokenizer is available)	COMMENT
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	394	# Step 6: Compare with transformers tokenizer if available	COMMENT
LOW	tests/unit_tests/cpu/test_tokenizer.py	351	# Step 1: Download tokenizer files	COMMENT
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	103	0.5, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	104	1.0, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	105	1.0, # Step 2: We maunally added step of stable phase, to prevent LR from dropping to 0 at last step	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	106	7.0 / 8.0, # Step 3: 7/8 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	107	6.0 / 8.0, # Step 4: 3/4 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	108	5.0 / 8.0, # Step 5: 5/8 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	109	4.0 / 8.0, # Step 6: 1/2 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	110	3.0 / 8.0, # Step 7: 3/8 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	111	2.0 / 8.0, # Step 8: 1/4 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	112	1.0 / 8.0, # Step 9: 1/8 of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	146	0.5, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	147	1.0, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	148	1.0, # Step 2: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	149	1.0, # Step 3: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	150	1.0, # Step 4: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	151	1.0, # Step 5: We maunally added step of stable phase, to prevent LR from dropping to 0 at last step	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	152	0.8, # Step 6: Linear decay starts (80% of max LR)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	153	0.6, # Step 7: 60% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	154	0.4, # Step 8: 40% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	155	0.2, # Step 9: 20% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	212	0.2, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	213	0.4, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	214	0.6, # Step 2: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	215	0.8, # Step 3: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	216	1.0, # Step 4: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	249	0.5, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	250	1.0, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	251	1.0, # Step 2: We maunally added step of stable phase, to prevent LR from dropping to 0 at last step	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	252	1.0, # Step 3: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	253	1.0, # Step 4: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	254	1.0, # Step 5: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	255	1.0, # Step 6: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	256	1.0, # Step 7: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	257	1.0, # Step 8: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	258	1.0, # Step 9: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	292	0.2, # Step 0: 50% of max LR (warmup)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	293	0.4, # Step 1: 100% of max LR (warmup complete)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	294	0.6, # Step 2: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	295	0.8, # Step 3: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	296	1.0, # Step 4: Stable phase	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	297	1.0, # Step 5: We maunally added step of stable phase, to prevent LR from dropping to 0 at last step	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	298	0.8, # Step 6: Linear decay starts (80% of max LR)	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	299	0.6, # Step 7: 60% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	300	0.4, # Step 8: 40% of max LR	CODE
LOW⚡	tests/unit_tests/cpu/test_lr_scheduler.py	301	0.2, # Step 9: 20% of max LR	CODE
LOW⚡	scripts/checkpoint_compat_test.py	228	# Step 1: Reference run at load_commit (full training from scratch)	COMMENT
LOW⚡	scripts/checkpoint_compat_test.py	237	# Step 2: Save run at save_commit (partial training + checkpoint)	COMMENT
LOW	scripts/checkpoint_compat_test.py	253	# Step 3: Resume run at load_commit (load checkpoint, train to end)	COMMENT
LOW	scripts/checkpoint_compat_test.py	268	# Step 4: Compare	COMMENT
LOW⚡	.github/workflows/integration_test_8gpu_torchft.yaml	32	# Step 1: Dynamically compute the matrix based on conditions	COMMENT
LOW⚡	.github/workflows/integration_test_8gpu_torchft.yaml	40	# Step 2: Use the dynamic matrix in the build-test job	COMMENT
LOW⚡	.github/workflows/integration_test_8gpu_rl.yaml	30	# Step 1: Dynamically compute the matrix based on conditions	COMMENT
LOW⚡	.github/workflows/integration_test_8gpu_rl.yaml	40	# Step 2: Use the dynamic matrix in the build-test job	COMMENT
LOW⚡	…tegration_test_8gpu_transformers_modeling_backend.yaml	31	# Step 1: Dynamically compute the matrix based on conditions	COMMENT
27 more matches not shown…

Modern Structural Boilerplate111 hits · 112 pts

Severity	File	Line	Snippet	Context
LOW	tests/integration_tests/__init__.py	12	__all__ = [	CODE
LOW⚡	…skills/numerics_debugging/scripts/activation_tracer.py	377	def set_numerics_capture_active(active: bool) -> None:	CODE
LOW	…skills/numerics_debugging/scripts/activation_tracer.py	873	def _setup(self) -> None:	CODE
LOW	torchtitan/overrides/fused_mla.py	77	__all__ = [	CODE
LOW	torchtitan/overrides/helion_rope.py	69	__all__ = ["HelionComplexRoPE", "HelionCosSinRoPE", "HelionRoPE"]	CODE
LOW	torchtitan/overrides/fused_swiglu.py	79	__all__ = [	CODE
LOW	torchtitan/distributed/spmd_types.py	33	__all__ = [	CODE
LOW	torchtitan/distributed/__init__.py	9	__all__ = [	CODE
LOW	torchtitan/distributed/utils.py	39	def set_spmd_backend(spmd_backend: str) -> None:	CODE
LOW	torchtitan/distributed/utils.py	318	def set_batch_invariance(enable: bool) -> None:	CODE
LOW	torchtitan/distributed/parallel_dims.py	21	__all__ = [	CODE
LOW	torchtitan/distributed/pipeline_parallel.py	38	__all__ = ["pipeline_llm", "pipeline_vlm"]	CODE
LOW	torchtitan/distributed/flex_shard/dist_muon.py	53	__all__ = [	CODE
LOW	torchtitan/distributed/flex_shard/__init__.py	12	__all__ = [	CODE
LOW	torchtitan/distributed/flex_shard/optimizer_reshard.py	18	__all__ = ["BlockShard", "BucketConfig", "ComputeLayout", "Owned"]	CODE
LOW	torchtitan/distributed/deepep/hybridep.py	543	__all__ = [	CODE
LOW	torchtitan/distributed/context_parallel/__init__.py	21	__all__ = [	CODE
LOW	torchtitan/distributed/minimal_async_ep/__init__.py	15	__all__ = [	CODE
LOW	torchtitan/distributed/minimal_async_ep/api.py	1017	__all__ = [	CODE
LOW	torchtitan/config/configurable.py	15	logger = logging.getLogger(__name__)	CODE
LOW	torchtitan/config/__init__.py	34	__all__ = [	CODE
LOW	…/experiments/transformers_modeling_backend/__init__.py	16	__all__ = [	CODE
LOW	…periments/transformers_modeling_backend/hf_sharding.py	242	def _set_layer_sharding_configs(layer: nn.Module, *, enable_sp: bool) -> None:	CODE
LOW	torchtitan/experiments/graph_trainer/custom_codegen.py	418	def _setup_custom_codegen(self) -> None:	CODE
LOW	torchtitan/experiments/graph_trainer/ep_chunk_pass.py	185	def _set_direction_meta(node: fx.Node, *, is_backward: bool) -> None:	CODE
LOW	torchtitan/experiments/graph_trainer/ep_overlap_pass.py	159	def _set_copy_non_blocking(node: fx.Node, non_blocking: bool) -> None:	CODE
LOW	…ents/graph_trainer/tests/test_bitwise_deterministic.py	71	def _set_deterministic(seed: int = SEED) -> None:	CODE
LOW	…xperiments/graph_trainer/tests/test_sac_peak_memory.py	30	def _set_deterministic() -> None:	CODE
LOW	…xperiments/graph_trainer/tests/test_graph_pp_passes.py	419	def _setup(self) -> None:	CODE
LOW	torchtitan/experiments/graph_trainer/graph_pp/runner.py	46	__all__ = [	CODE
LOW	…chtitan/experiments/graph_trainer/graph_pp/__init__.py	25	__all__ = [	CODE
LOW	torchtitan/experiments/torchft/__init__.py	14	__all__ = [	CODE
LOW	torchtitan/experiments/torchft/optimizer.py	19	__all__ = ["TorchFTOptimizersContainer"]	CODE
LOW	torchtitan/experiments/torchft/config/__init__.py	10	__all__ = [	CODE
LOW	…chtitan/experiments/torchft/tests/integration_tests.py	16	logger = logging.getLogger(__name__)	CODE
LOW	torchtitan/experiments/torchft/diloco/__init__.py	9	__all__ = [	CODE
LOW	torchtitan/experiments/forge/__init__.py	9	__all__ = ["ForgeEngine"]	CODE
LOW	torchtitan/experiments/rl/controller.py	140	logger = logging.getLogger(__name__)	CODE
LOW	torchtitan/experiments/rl/renderer.py	16	logger = logging.getLogger(__name__)	CODE
LOW	torchtitan/experiments/rl/__init__.py	47	__all__ = [	CODE
LOW	torchtitan/experiments/rl/train.py	38	logger = logging.getLogger(__name__)	CODE
LOW	torchtitan/experiments/rl/batch_invariance.py	22	logger = logging.getLogger(__name__)	CODE
LOW	torchtitan/experiments/rl/losses/__init__.py	10	__all__ = ["DAPOLoss", "GRPOLoss"]	CODE
LOW⚡	torchtitan/experiments/rl/tests/test_bitwise_parity.py	93	logger = logging.getLogger(__name__)	CODE
LOW	torchtitan/experiments/rl/tests/test_bitwise_parity.py	188	def _set_generator_determinism(debug) -> None:	CODE
LOW	torchtitan/experiments/rl/models/attention.py	34	logger = logging.getLogger(__name__)	CODE
LOW	…titan/experiments/rl/observability/metrics/__init__.py	14	__all__ = [	CODE
LOW	…itan/experiments/rl/observability/metrics/processor.py	30	__all__ = [	CODE
LOW	…rchtitan/experiments/rl/observability/metrics/types.py	17	__all__ = [	CODE
LOW	torchtitan/experiments/rl/rollout/rollouter.py	39	logger = logging.getLogger(__name__)	CODE
LOW	torchtitan/experiments/rl/rollout/__init__.py	17	__all__ = [	CODE
LOW	torchtitan/experiments/rl/components/batcher.py	26	logger = logging.getLogger(__name__)	CODE
LOW	torchtitan/experiments/rl/rubrics/__init__.py	9	__all__ = ["RewardFn", "Rubric", "RubricOutput"]	CODE
LOW	torchtitan/experiments/rl/examples/search_r1/rubric.py	127	__all__ = ["RewardExactMatch"]	CODE
LOW	torchtitan/experiments/rl/examples/search_r1/env.py	26	logger = logging.getLogger(__name__)	CODE
LOW	…rchtitan/experiments/rl/examples/search_r1/__init__.py	15	__all__ = [	CODE
LOW	…htitan/experiments/rl/examples/alphabet_sort/rubric.py	116	__all__ = ["RewardAlphabetSort", "score_sorted_list"]	STRING
LOW	…itan/experiments/rl/examples/alphabet_sort/__init__.py	18	__all__ = [	CODE
LOW	…rchtitan/experiments/rl/examples/dapo_math/__init__.py	19	__all__ = [	CODE
LOW	torchtitan/experiments/rl/actors/generator.py	69	logger = logging.getLogger(__name__)	CODE
51 more matches not shown…

Docstring Block Structure18 hits · 90 pts

Severity	File	Line	Snippet	Context
HIGH	…h_trace_compaction/scripts/compact_cuda_graph_trace.py	182	Extract measured GPU slices that have a profiler stream. Args: events: Trace event list. Returns:	STRING
HIGH	…h_trace_compaction/scripts/compact_cuda_graph_trace.py	463	Verify slice identity and non-overlap of every compacted lane. Args: original_events: Unmodified trace even	STRING
HIGH	…h_trace_compaction/scripts/compact_cuda_graph_trace.py	747	Remap GPU flows and annotations, and replace old stream metadata. Args: events: Events whose measured slice	STRING
HIGH	…h_trace_compaction/scripts/compact_cuda_graph_trace.py	806	Compact CUDA graph streams in a parsed PyTorch profiler trace. Args: trace: Parsed profiler trace. Ret	STRING
HIGH	torchtitan/distributed/utils.py	580	Clip the gradient norm of an iterable of parameters. Gradient norm clipping requires computing the gradient no	STRING
HIGH	torchtitan/distributed/parallel_dims.py	337	Get a device mesh by dimension name(s), returning None if not enabled. Args: dims: Names of the mes	STRING
HIGH	torchtitan/distributed/parallel_dims.py	400	Get a device mesh by dimension name(s), raising if not available. Args: dims: Names of the mesh dim	STRING
HIGH	torchtitan/distributed/pipeline_parallel.py	580	Create pipeline stages based on specified module names for each stage. Also used by Graph PP to split the model int	STRING
HIGH	torchtitan/distributed/context_parallel/api.py	133	Shard inputs and attention masks across the context parallel mesh. This function distributes input tensors acr	STRING
HIGH	…/experiments/transformers_modeling_backend/pipeline.py	156	This API creates pipeline stages based on specified module names for each stage. Some model restrictions inclu	STRING
HIGH	…itan/experiments/graph_trainer/graph_pp/split_di_dw.py	156	Split a backward graph into input-gradient and weight-gradient graphs. Contract: Input: backward(save	STRING
HIGH	torchtitan/experiments/graph_trainer/graph_pp/runner.py	714	Register GraphPP action handlers on a runtime PP schedule. Args: schedule (_PipelineScheduleRuntime): Runti	STRING
HIGH	…htitan/experiments/graph_trainer/graph_pp/partition.py	445	Partition a post-pass GraphTrainer joint graph into GraphPP callables. The caller passes a flat ``minimal_fx_tracer	STRING
HIGH	…ments/graph_trainer/graph_pp/split_fsdp_collectives.py	95	Split forward FSDP all-gather chains from a forward graph. Contract: unshard(param_shards_and_replicated_para	STRING
HIGH	…ments/graph_trainer/graph_pp/split_fsdp_collectives.py	251	Split backward FSDP/DDP/HSDP reduce-grad epilogues. Contract: bw_no_fsdp(original_backward_inputs) ->	STRING
HIGH	…/experiments/graph_trainer/graph_pp/graph_multiplex.py	48	Concatenate backward and forward graphs into one boxed GraphPP callable. Contract: ``OVERLAP_F_B`` schedule a	STRING
HIGH	torchtitan/experiments/torchft/diloco/utils.py	19	This API creates fragments based on specified module names for each fragment. This method updates the model in	STRING
HIGH	torchtitan/models/common/multimodal.py	43	Locate each visual item's placeholder run in the token sequence. Args: tokens: ``(T,)`` token IDs.	STRING

Self-Referential Comments26 hits · 84 pts

Severity	File	Line	Snippet	Context
MEDIUM	tests/unit_tests/cpu/test_set_determinism.py	46	# Create a world_mesh mock	COMMENT
MEDIUM⚡	tests/unit_tests/cpu/test_lr_scheduler.py	27	# Create a simple model with parameters	COMMENT
MEDIUM⚡	tests/unit_tests/cpu/test_lr_scheduler.py	29	# Create an optimizer	COMMENT
MEDIUM⚡	tests/unit_tests/cpu/test_lr_scheduler.py	36	# Create an optimizer container	COMMENT
MEDIUM	tests/unit_tests/cpu/test_lr_scheduler.py	49	# Create a trainer config with the specified parameters	COMMENT
MEDIUM	tests/unit_tests/cpu/test_lr_scheduler.py	86	# Create a job config with 10 steps, 2 warmup steps, and linear decay	COMMENT
MEDIUM	tests/unit_tests/cpu/test_lr_scheduler.py	129	# Create a job config with 10 steps, 2 warmup steps, 3 stable steps, and 5 decay steps	COMMENT
MEDIUM	tests/unit_tests/cpu/test_lr_scheduler.py	171	# Create a job config with a minimum learning rate	COMMENT
MEDIUM	tests/unit_tests/cpu/test_lr_scheduler.py	195	# Create a job config where warmup steps > training steps	COMMENT
MEDIUM	tests/unit_tests/cpu/test_lr_scheduler.py	232	# Create a job config with 10 steps, 2 warmup steps, and no decay phase	COMMENT
MEDIUM	tests/unit_tests/cpu/test_lr_scheduler.py	274	# Create a job config where warmup + decay steps > training steps	COMMENT
MEDIUM	tests/unit_tests/cpu/test_checkpoint.py	772	# Create a fake model with freqs_cis and other parameters	COMMENT
MEDIUM	tests/assets/cc12m_test/pack_test_dataset.py	23	# Create a TarWriter object to write the dataset to a tar archive	COMMENT
MEDIUM⚡	scripts/loss_compare.py	350	# Create the output folder	COMMENT
MEDIUM⚡	scripts/loss_compare.py	845	# Create a test case	COMMENT
MEDIUM	…ripts/checkpoint_conversion/numerical_tests_example.py	151	# Define the set of outputs to test loss for	COMMENT
MEDIUM	.github/scripts/update_version.sh	6	# Create a variable named __version__ at the end of __init__.py	COMMENT
MEDIUM	torchtitan/distributed/activation_checkpoint.py	7	# This file provides the util functions to apply activation checkpointing to the model.	COMMENT
MEDIUM	torchtitan/distributed/pipeline_parallel.py	489	# Create a set of modules to keep for faster lookup	COMMENT
MEDIUM	…/experiments/transformers_modeling_backend/pipeline.py	196	# Create a set of modules to keep for faster lookup	COMMENT
MEDIUM	…chtitan/experiments/graph_trainer/tests/test_passes.py	389	# Create a second PG to simulate expert-FSDP.	COMMENT
MEDIUM	…experiments/graph_trainer/tests/test_custom_codegen.py	206	# Create main module	COMMENT
MEDIUM	torchtitan/experiments/torchft/manager.py	199	# Create the outer optimizer based on the inner optimizer parameters.	COMMENT
MEDIUM	torchtitan/experiments/torchft/diloco/utils.py	48	# Create a set of modules to keep for faster lookup	COMMENT
MEDIUM	torchtitan/experiments/rl/environment/token.py	226	# Create a new list to avoid mutating previous states	COMMENT
MEDIUM	torchtitan/models/common/moe.py	184	"""This class implements token-choice routing. In token-choice top-K routing, each token is	STRING

Excessive Try-Catch Wrapping58 hits · 70 pts

Severity	File	Line	Snippet	Context
LOW	tests/unit_tests/cpu/test_tp_kv_heads_validation.py	38	except Exception:	CODE
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	376	except Exception as e:	CODE
LOW⚡	tests/unit_tests/cpu/test_tokenizer.py	385	except Exception as e:	CODE
LOW⚡	tests/unit_tests/cpu/test_checkpoint.py	957	except Exception as e:	CODE
LOW	tests/integration_tests/run_tests.py	456	except Exception as e:	CODE
LOW	tests/integration_tests/run_tests.py	468	except Exception as e:	CODE
MEDIUM⚡	scripts/loss_compare.py	325	log_print(f"Error: Import file does not exist: {import_result}")	CODE
MEDIUM⚡	scripts/loss_compare.py	346	log_print(f"Error: Output folder '{output_folder}' already exists")	CODE
MEDIUM⚡	scripts/loss_compare.py	834	log_print(f"Error: No {metric_name} values found in baseline")	CODE
MEDIUM⚡	scripts/loss_compare.py	838	log_print(f"Error: No {metric_name} values found in test")	CODE
MEDIUM⚡	scripts/loss_compare.py	842	log_print(f"Error: No {metric_name} values found in import file")	CODE
MEDIUM	scripts/loss_compare.py	290	log_print("Error: All settings are identical")	CODE
MEDIUM	scripts/loss_compare.py	303	log_print(f"Error: --steps must be a positive integer, got: {steps}")	CODE
MEDIUM	scripts/loss_compare.py	308	log_print("Error: --import-result requires --assert-equal")	CODE
MEDIUM	scripts/loss_compare.py	788	log_print("Error: baseline-only mode requires --import-result")	CODE
MEDIUM	scripts/loss_compare.py	799	log_print(f"Error: metric '{name}' missing from {other_name} results")	CODE
LOW	.github/scripts/todo_debt_tracker.py	45	except Exception:	CODE
LOW	.github/scripts/todo_debt_tracker.py	116	except Exception as e:	CODE
LOW	.github/scripts/todo_debt_tracker.py	183	except Exception:	CODE
LOW	.github/scripts/todo_debt_tracker.py	303	except Exception as e:	CODE
MEDIUM	.github/scripts/todo_debt_tracker.py	304	print(f"Error reading {file_path}: {e}", file=sys.stderr)	CODE
LOW	torchtitan/__init__.py	11	except Exception:	CODE
LOW	torchtitan/train.py	59	except Exception:	CODE
LOW	…tan/experiments/transformers_modeling_backend/model.py	741	except Exception as e:	CODE
LOW	…_moe_model/scripts/numerical_equivalence_pretrained.py	47	except Exception:	CODE
LOW	…_moe_model/scripts/numerical_equivalence_pretrained.py	192	except Exception as e:	CODE
LOW	…/skills/add_moe_model/scripts/numerical_equivalence.py	572	except Exception as e:	CODE
LOW	…/skills/add_moe_model/scripts/numerical_equivalence.py	622	except Exception as e:	CODE
LOW	torchtitan/experiments/graph_trainer/custom_codegen.py	64	except Exception:	CODE
LOW	torchtitan/experiments/graph_trainer/ep_pass_utils.py	218	except Exception:	CODE
LOW	torchtitan/experiments/graph_trainer/cpu_offload.py	167	except Exception:	CODE
LOW	torchtitan/experiments/graph_trainer/cpu_offload.py	179	except Exception:	CODE
LOW	…eriments/graph_trainer/log_activation_memory_policy.py	73	except Exception:	CODE
LOW	torchtitan/experiments/graph_trainer/graph_pp/utils.py	221	except Exception as exc:	CODE
LOW	torchtitan/experiments/graph_trainer/graph_pp/utils.py	241	except Exception as exc:	CODE
LOW	torchtitan/experiments/forge/example_train.py	457	except Exception:	CODE
LOW	torchtitan/experiments/rl/controller.py	453	except Exception:	CODE
LOW	torchtitan/experiments/rl/controller.py	458	except Exception:	CODE
LOW⚡	torchtitan/experiments/rl/controller.py	476	except Exception:	CODE
LOW⚡	torchtitan/experiments/rl/controller.py	481	except Exception:	CODE
LOW⚡	torchtitan/experiments/rl/controller.py	487	except Exception:	CODE
LOW	torchtitan/experiments/rl/controller.py	984	except Exception:	CODE
LOW	…/experiments/rl/observability/vllm_otel_stat_logger.py	269	except Exception as e:	CODE
LOW	…/experiments/rl/observability/vllm_otel_stat_logger.py	374	except Exception as e:	CODE
LOW	…itan/experiments/rl/observability/metrics/processor.py	212	except Exception:	CODE
LOW	…itan/experiments/rl/observability/metrics/processor.py	223	except Exception:	CODE
LOW	torchtitan/experiments/rl/rollout/rollouter.py	435	except Exception:	CODE
LOW	torchtitan/experiments/rl/examples/search_r1/env.py	90	except Exception as exc:	CODE
LOW	torchtitan/experiments/rl/actors/generator.py	1218	except Exception as exc:	CODE
LOW	torchtitan/experiments/rl/actors/generator.py	1412	except Exception:	CODE
MEDIUM	torchtitan/experiments/rl/scripts/loss_compare.py	293	log_print("Error: no losses found in TensorBoard events")	CODE
LOW	torchtitan/experiments/rl/environment/token.py	165	except Exception:	CODE
LOW	torchtitan/observability/structured_logger/README.md	147	except Exception:	CODE
LOW	torchtitan/components/metrics.py	438	except Exception as e:	CODE
LOW	torchtitan/components/data/loader.py	173	except Exception:	CODE
LOW	torchtitan/components/checkpointer/base.py	58	except Exception as error:	CODE
LOW	torchtitan/hf_datasets/multimodal/utils/video.py	93	except Exception as e:	CODE
LOW	torchtitan/hf_datasets/multimodal/utils/image.py	275	except Exception as e:	CODE

Over-Commented Block58 hits · 58 pts

Severity	File	Line	Snippet	Context
LOW	run_train.sh	1	#!/usr/bin/bash	COMMENT
LOW	.ci/docker/common/utils.sh	1	#!/bin/bash	COMMENT
LOW	.ci/docker/common/install_deepep.sh	1	#!/bin/bash	COMMENT
LOW	tests/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	tests/unit_tests/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	tests/unit_tests/gpu/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	tests/unit_tests/gpu/flex_shard/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	tests/unit_tests/cpu/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	tests/unit_tests/cpu/flex_shard/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	tests/unit_tests/cpu/observability/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	…skills/numerics_debugging/scripts/activation_tracer.py	81	on ``input_hashes`` / ``output_hash`` for comparison.	COMMENT
LOW	…skills/numerics_debugging/scripts/activation_tracer.py	201	# Indexing	COMMENT
LOW	…skills/numerics_debugging/scripts/activation_tracer.py	561	record_ids=True,	COMMENT
LOW	…skills/numerics_debugging/scripts/activation_tracer.py	641	# through to the metadata block below.	COMMENT
LOW	…/skills/numerics_debugging/scripts/compare_numerics.py	61	stats: dict[str, str] = field(default_factory=dict)	COMMENT
LOW	…/skills/numerics_debugging/scripts/compare_numerics.py	181		COMMENT
LOW	…/skills/numerics_debugging/scripts/compare_numerics.py	661	# log_tensor_hashes. Comma-separated lists of floats — we	COMMENT
LOW	.github/workflows/set-matrix.yaml	81	# Temporarily disable ROCm CI while the ROCm jobs are red.	COMMENT
LOW	.github/workflows/set-matrix.yaml	101	# "index-url": "https://download.pytorch.org/whl/nightly/rocm7.2",	COMMENT
LOW	.github/scripts/run_8xgpu_integration_tests.sh	1	#!/bin/bash	COMMENT
LOW	torchtitan_recipes/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/tools/utils.py	201	# data from https://www.amd.com/en/products/accelerators/instinct/mi200/mi250x.html (per GCD)	COMMENT
LOW	torchtitan/distributed/linear.py	161	# AG(X_k.T) @ dY produces dW.T. This mirrors the usual AG-linear wgrad	COMMENT
LOW	torchtitan/distributed/deepep/deepep.py	81	# ElasticBuffer.dispatch/combine are not autograd-aware. We wrap them in	COMMENT
LOW	torchtitan/config/manager.py	281	)	COMMENT
LOW	…tan/experiments/transformers_modeling_backend/model.py	741	except Exception as e:	COMMENT
LOW	…rmers_modeling_backend/tests/test_flex_cp_numerical.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	…iments/transformers_modeling_backend/tests/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	…s/transformers_modeling_backend/tests/run_moe_tests.sh	1	#!/bin/bash	COMMENT
LOW	torchtitan/experiments/graph_trainer/make_fx_tracer.py	481	# engine dispatches backward to a worker thread that has a	COMMENT
LOW	torchtitan/experiments/graph_trainer/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	…itan/experiments/graph_trainer/run_train_precompile.sh	1	#!/usr/bin/bash	COMMENT
LOW	torchtitan/experiments/graph_trainer/inductor_passes.py	201	raise ValueError(	COMMENT
LOW	torchtitan/experiments/graph_trainer/tests/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	…n/experiments/graph_trainer/tests/integration_tests.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	…n/experiments/graph_trainer/tests/integration_tests.py	721		COMMENT
LOW	torchtitan/experiments/torchft/tests/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/experiments/rl/controller.py	641	)	COMMENT
LOW	torchtitan/experiments/rl/tests/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/experiments/rl/tests/integration_tests.py	121	),	COMMENT
LOW	torchtitan/experiments/rl/models/attention.py	141	# absent). Re-read live per-layer metadata + kv_cache from the forward context	COMMENT
LOW	torchtitan/experiments/rl/models/vllm_wrapper.py	381	)	COMMENT
LOW	torchtitan/experiments/rl/observability/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/experiments/rl/components/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/experiments/rl/examples/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/experiments/rl/actors/generator.py	1341	# state_dict() returns hook-produced copies for fused modules (e.g.	COMMENT
LOW	torchtitan/experiments/rl/scripts/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/models/kimi_k2_7/config_registry.py	521	# Fail during config parsing, before TP/FSDP creates _StridedShard	COMMENT
LOW	torchtitan/models/flux/model/model.py	81	# Correction 1: DoubleStreamBlocks have symmetric img/txt streams;	COMMENT
LOW	torchtitan/models/common/attention.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/models/muse_glimmer/model.py	441	with multimodal_context():	COMMENT
LOW	torchtitan/observability/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/components/optimizer/lr_scheduler.py	221		COMMENT
LOW	torchtitan/components/quantization/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/components/quantization/nvfp4.py	161	self,	COMMENT
LOW	torchtitan/hf_datasets/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/hf_datasets/multimodal/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT
LOW	torchtitan/ops/__init__.py	1	# Copyright (c) Meta Platforms, Inc. and affiliates.	COMMENT

Fake / Example Data56 hits · 51 pts

Severity	File	Line	Snippet	Context
LOW	torchtitan/experiments/graph_trainer/fsdp_passes.py	91	for placeholder in gm.graph.find_nodes(op="placeholder"):	CODE
LOW	torchtitan/experiments/graph_trainer/fsdp_passes.py	130	if n.all_input_nodes[0].op == "placeholder":	CODE
LOW	torchtitan/experiments/graph_trainer/fsdp_passes.py	1120	if n in visited or n.op in ("placeholder", "get_attr"):	CODE
LOW	torchtitan/experiments/graph_trainer/ep_pass_utils.py	312	or node.op != "placeholder"	CODE
LOW	torchtitan/experiments/graph_trainer/ep_pass_utils.py	329	if node.op != "placeholder" or (val := tensor_meta(node)) is None:	CODE
LOW	torchtitan/experiments/graph_trainer/memory_policy.py	70	for node in gm.graph.find_nodes(op="placeholder"):	CODE
LOW⚡	torchtitan/experiments/graph_trainer/inductor_passes.py	64	if node.op == "placeholder" and "val" in node.meta:	CODE
LOW⚡	torchtitan/experiments/graph_trainer/inductor_passes.py	80	if node.op == "placeholder":	CODE
LOW⚡	torchtitan/experiments/graph_trainer/inductor_passes.py	101	original_placeholders = list(gm.graph.find_nodes(op="placeholder"))	CODE
LOW⚡	torchtitan/experiments/graph_trainer/inductor_passes.py	110	if node.op == "placeholder":	CODE
LOW	torchtitan/experiments/graph_trainer/inductor_passes.py	383	if node.op in ("placeholder", "output"):	CODE
LOW	torchtitan/experiments/graph_trainer/ep_chunk_pass.py	573	if node.op != "placeholder":	CODE
LOW	torchtitan/experiments/graph_trainer/ep_chunk_pass.py	2315	for idx, node in enumerate(n for n in gm.graph.nodes if n.op == "placeholder")	CODE
LOW⚡	torchtitan/experiments/graph_trainer/ep_overlap_pass.py	81	_GRAPH_BOUNDARY_OPS = {"placeholder", "get_attr"}	CODE
LOW	torchtitan/experiments/graph_trainer/debug_utils.py	34	if node.op == "placeholder":	CODE
LOW	…n/experiments/graph_trainer/tests/test_trace_module.py	89	if node.op != "placeholder":	CODE
LOW	…n/experiments/graph_trainer/tests/test_trace_module.py	172	if node.op == "placeholder"	CODE
LOW	…n/experiments/graph_trainer/tests/test_trace_module.py	427	if node.op == "placeholder"	CODE
LOW	…n/experiments/graph_trainer/tests/test_trace_module.py	451	if node.op == "placeholder"	CODE
LOW	…n/experiments/graph_trainer/tests/test_trace_module.py	472	if node.op == "placeholder"	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	2132	self.assertEqual(node.args[0].op, "placeholder")	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	5481	if node.op == "placeholder":	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	5588	if node.op == "placeholder":	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	5607	self.assertEqual(node.args[0].op, "placeholder")	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	5685	self.assertEqual(relu_node.args[0].op, "placeholder")	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	5784	if node.op == "placeholder":	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	5884	if node.op == "placeholder":	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	5925	if node.op == "placeholder":	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	6255	node.target for node in compile_gm.graph.find_nodes(op="placeholder")	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	6258	node.meta for node in compile_gm.graph.find_nodes(op="placeholder")	CODE
LOW	…chtitan/experiments/graph_trainer/tests/test_passes.py	6295	node.target for node in wrapped.graph.find_nodes(op="placeholder")	CODE
LOW	…xperiments/graph_trainer/tests/test_graph_pp_runner.py	479	for node in gm.graph.find_nodes(op="placeholder"):	CODE
LOW	…xperiments/graph_trainer/tests/test_graph_pp_runner.py	636	bw_gm.graph.find_nodes(op="placeholder")[0].meta["val"] = bw_fake	CODE
LOW	…xperiments/graph_trainer/tests/test_graph_pp_runner.py	637	fw_gm.graph.find_nodes(op="placeholder")[0].meta["val"] = fw_fake	CODE
LOW	…xperiments/graph_trainer/tests/test_graph_pp_runner.py	652	placeholders = multiplexed.graph.find_nodes(op="placeholder")	CODE
LOW	…xperiments/graph_trainer/tests/test_graph_pp_runner.py	696	fw_gm.graph.find_nodes(op="placeholder")[0].meta["val"] = fw_fake	CODE
LOW⚡	…xperiments/graph_trainer/tests/test_graph_pp_passes.py	659	sharded_param = gm.graph.find_nodes(op="placeholder")[0]	CODE
LOW	…xperiments/graph_trainer/tests/test_graph_pp_passes.py	534	return tuple(node.name for node in gm.graph.find_nodes(op="placeholder"))	CODE
LOW	…itan/experiments/graph_trainer/graph_pp/split_di_dw.py	200	for placeholder in list(bw_gm.graph.find_nodes(op="placeholder")):	CODE
LOW	…itan/experiments/graph_trainer/graph_pp/split_di_dw.py	215	placeholders = list(bw_gm.graph.find_nodes(op="placeholder"))	CODE
LOW⚡	…htitan/experiments/graph_trainer/graph_pp/partition.py	197	for node in gm.graph.find_nodes(op="placeholder"):	CODE
LOW	…htitan/experiments/graph_trainer/graph_pp/partition.py	320	if mutated_base.op != "placeholder" and mutated_base in backward_nodes:	CODE
LOW	…htitan/experiments/graph_trainer/graph_pp/partition.py	348	and node.op == "placeholder"	CODE
LOW	…htitan/experiments/graph_trainer/graph_pp/partition.py	475	placeholders = list(joint.graph.find_nodes(op="placeholder"))	CODE
LOW	torchtitan/experiments/graph_trainer/graph_pp/utils.py	289	return tuple(node.name for node in gm.graph.find_nodes(op="placeholder"))	CODE
LOW	torchtitan/experiments/graph_trainer/graph_pp/utils.py	322	return {node for node in node_closure(values) if node.op == "placeholder"}	CODE
LOW⚡	torchtitan/experiments/graph_trainer/graph_pp/utils.py	386	if node.op != "placeholder":	CODE
LOW⚡	torchtitan/experiments/graph_trainer/graph_pp/utils.py	399	for node in gm.graph.find_nodes(op="placeholder"):	CODE
LOW	…an/experiments/graph_trainer/graph_pp/graph_builder.py	384	placeholders = self.modules.fw.graph.find_nodes(op="placeholder")	CODE
LOW	…an/experiments/graph_trainer/graph_pp/graph_builder.py	750	if node.op == "placeholder":	CODE
LOW	…ments/graph_trainer/graph_pp/split_fsdp_collectives.py	132	placeholders = graph.find_nodes(op="placeholder")	CODE
LOW	…ments/graph_trainer/graph_pp/split_fsdp_collectives.py	288	placeholders = graph.find_nodes(op="placeholder")	CODE
LOW	…/experiments/graph_trainer/graph_pp/graph_multiplex.py	104	fw_placeholders = multiplexed_gm.graph.find_nodes(op="placeholder")	CODE
LOW	…/experiments/graph_trainer/graph_pp/graph_multiplex.py	109	for node in bw_gm.graph.find_nodes(op="placeholder"):	CODE
LOW	…/experiments/graph_trainer/graph_pp/graph_multiplex.py	121	(node for node in multiplexed_gm.graph.nodes if node.op != "placeholder"),	CODE
LOW	…/experiments/graph_trainer/graph_pp/graph_multiplex.py	128	bw_nodes = dropwhile(lambda node: node.op == "placeholder", bw_nodes)	CODE

Cross-Language Confusion6 hits · 29 pts

Severity	File	Line	Snippet	Context
HIGH	tests/unit_tests/cpu/test_override.py	477	parse_cli_imports(['mod={"block_size": 256, "flag": null}']),	CODE
HIGH	…/skills/numerics_debugging/scripts/compare_numerics.py	1016	const hidden = el.dataset.hideByCategory \|\| el.dataset.hideByOpFilter;	CODE
HIGH	…/skills/numerics_debugging/scripts/compare_numerics.py	1028	let _opFilterTimer = null;	CODE
HIGH	…/skills/numerics_debugging/scripts/compare_numerics.py	1037	let re = null;	CODE
HIGH	scripts/loss_compare.py	482	log_print(" - To commit: git add -A && git commit -m 'message'")	CODE
HIGH⚡	…ripts/checkpoint_conversion/numerical_tests_qwen3_5.py	53	"""KL(softmax(b) \|\| softmax(a)) — F.kl_div(log Q, P) computes KL(P \|\| Q)."""	STRING

Redundant / Tautological Comments18 hits · 24 pts

Severity	File	Line	Snippet	Context
LOW	run_train.sh	19	# Set RANK to simulate a nonzero global rank, for example RANK=16.	COMMENT
LOW	tests/unit_tests/cpu/test_checkpoint.py	113	# Check if this is async_with_pinned_mem mode by looking for async_stager parameter	COMMENT
LOW	tests/unit_tests/cpu/test_checkpoint.py	713	cfg.interval = 10 # Set interval to 10 so step 1 wouldn't normally trigger save	CODE
LOW	tests/assets/cc12m_test/pack_test_dataset.py	38	# Write the file and its metadata to the TarWriter	COMMENT
LOW⚡	scripts/loss_compare.py	344	# Check if output folder already exists	COMMENT
LOW	scripts/loss_compare.py	760	# Check if losses were extracted successfully	COMMENT
LOW	scripts/loss_compare.py	1241	# Check if git working directory is clean before switching commits	COMMENT
LOW	torchtitan/config/manager.py	221	# Check if we are using tokenizer.model, if so then we need to alert users to redownload the tokenizer	COMMENT
LOW	…chtitan/experiments/torchft/tests/integration_tests.py	105	# Check if we have enough GPUs	COMMENT
LOW	torchtitan/experiments/rl/generate.py	246	# Print results	COMMENT
LOW	torchtitan/experiments/rl/actors/generator.py	1106	# Add the request to the queue; the engine loop will admit + process it.	COMMENT
LOW	torchtitan/experiments/rl/actors/trainer.py	275	# Check if the requested context exceeds the model context length.	COMMENT
LOW	torchtitan/models/flux/validate.py	147	# Set model to eval mode	COMMENT
LOW	torchtitan/models/flux/model/state_dict_adapter.py	38	# Check if safetensors index file exists	COMMENT
LOW	torchtitan/components/metrics.py	219	# Check if the loss-visible rank is included in LOG_RANK environment variable	COMMENT
LOW	torchtitan/components/metrics.py	390	# Check if any logging backend is enabled	COMMENT
LOW	torchtitan/components/loss.py	602	# Check if it's training model or validation mode	COMMENT
LOW	torchtitan/components/validate.py	148	# Set model to eval mode	COMMENT

AI Slop Vocabulary6 hits · 18 pts

Severity	File	Line	Snippet	Context
MEDIUM	tests/unit_tests/gpu/test_helion_rope.py	31	# tests below stay robust if a sibling test (e.g. test_override.py) calls	COMMENT
MEDIUM	tests/unit_tests/cpu/test_tokenizer.py	178	# 2. Compare vocabularies with more comprehensive sampling	COMMENT
MEDIUM	tests/unit_tests/cpu/test_tokenizer.py	203	# Test a random sample of tokens (more comprehensive than just common words)	COMMENT
MEDIUM	torchtitan/experiments/rl/routing/strategies.py	189	# easy to implement, but not robust for all scenarios. Revisit with an	COMMENT
MEDIUM	torchtitan/experiments/rl/routing/strategies.py	190	# more robust approach.	COMMENT
MEDIUM	torchtitan/components/tokenizer.py	371	# Process added_tokens_decoder (comprehensive special token definitions)	COMMENT

Cross-File Repetition3 hits · 15 pts

Severity	File	Snippet	Context
HIGH	…htitan/experiments/graph_trainer/llama3/parallelize.py	token count {training.num_tokens_per_microbatch_per_dp_rank} must be divisible by the product of tp degree ({parallel_di	STRING
HIGH	…chtitan/experiments/graph_trainer/qwen3/parallelize.py	token count {training.num_tokens_per_microbatch_per_dp_rank} must be divisible by the product of tp degree ({parallel_di	STRING
HIGH	…n/experiments/graph_trainer/deepseek_v3/parallelize.py	token count {training.num_tokens_per_microbatch_per_dp_rank} must be divisible by the product of tp degree ({parallel_di	STRING

AI Structural Patterns13 hits · 12 pts

Severity	File	Line	Context
LOW	tests/unit_tests/gpu/test_qwen3_5_deltanet.py	217	CODE
LOW	torchtitan/distributed/flex_shard/dist_muon.py	271	CODE
LOW	…periments/transformers_modeling_backend/parallelize.py	232	CODE
LOW	…formers_modeling_backend/tests/test_moe_parallelism.py	79	CODE
LOW	…formers_modeling_backend/tests/test_moe_parallelism.py	115	CODE
LOW	torchtitan/experiments/graph_trainer/simple_fsdp.py	250	CODE
LOW	…ents/graph_trainer/tests/test_bitwise_deterministic.py	185	CODE
LOW	…experiments/graph_trainer/tests/_trainer_test_utils.py	23	CODE
LOW	torchtitan/models/llama3/model.py	53	CODE
LOW	torchtitan/models/common/decoder.py	308	CODE
LOW	torchtitan/models/muse_glimmer/__init__.py	267	CODE
LOW	torchtitan/models/muse_glimmer/model.py	190	CODE
LOW	torchtitan/hf_datasets/multimodal/utils/image.py	200	CODE

Modern AI Meta-Vocabulary1 hit · 3 pts

Severity	File	Line	Snippet	Context
MEDIUM	torchtitan/experiments/rl/examples/search_r1/README.md	1	# Search-R1: multi-turn retrieval-augmented GRPO	COMMENT

AI Response Leakage1 hit · 2 pts

Severity	File	Line	Snippet	Context
HIGH	torchtitan/experiments/rl/actors/generator.py	674	# The loop now executes the decision. In this example: STEP.	STRING

Example Usage Blocks1 hit · 2 pts

Severity	File	Line	Snippet	Context
LOW	…s/transformers_modeling_backend/tests/run_moe_tests.sh	4	# Usage:	COMMENT

Analysis Overview

What These Metrics Mean

Score History

Severity Breakdown

Directory Score Breakdown

Pattern Findings