alibaba/ROLL

17.3

Adjusted Score

17.3

Raw Score

100%

Time Factor

2026-07-14

Last Push

3.3K

Stars

Python

Language

156.3K

Lines of Code

842

Files

1.8K

Pattern Hits

2026-07-14

Scan Date

0.10

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 12HIGH 70MEDIUM 200LOW 1519

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 1801 distinct pattern matches across 21 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 Identifiers482 hits · 490 pts

Severity	File	Line	Snippet	Context
LOW	mcore_adapter/src/mcore_adapter/initialize.py	16	def is_distribute_initialized():	CODE
LOW	mcore_adapter/src/mcore_adapter/training_args.py	425	def allow_variable_seq_lengths(self):	CODE
LOW	mcore_adapter/src/mcore_adapter/utils.py	81	def is_mcore_version_greater_than(content: str):	CODE
LOW	mcore_adapter/src/mcore_adapter/utils.py	105	def is_transformers_version_greater_than(content: str):	CODE
LOW	mcore_adapter/src/mcore_adapter/patcher.py	31	def patch_torch_find_nd_overlapping_shards():	CODE
LOW	mcore_adapter/src/mcore_adapter/patcher.py	37	def _find_nd_overlapping_shards(shards: list[ShardMetadata], sharded_dims: list[int]) -> Optional[tuple[int, int]]:	CODE
LOW	mcore_adapter/src/mcore_adapter/patcher.py	85	def patch_torch_validate_global_plan():	CODE
LOW	mcore_adapter/src/mcore_adapter/patcher.py	181	def patch_megatron_preload_tensors_non_blocking(non_blocking: bool = False):	CODE
LOW	mcore_adapter/src/mcore_adapter/patcher.py	214	def patched_get_save_function_and_args(self):	CODE
LOW	mcore_adapter/src/mcore_adapter/checkpointing.py	18	def get_checkpoint_tracker_filename(checkpoints_path):	CODE
LOW	mcore_adapter/src/mcore_adapter/checkpointing.py	278	def load_state_dict_from_checkpoint(checkpoint_dir, models = None):	CODE
LOW	mcore_adapter/src/mcore_adapter/checkpointing.py	286	def save_config_and_state_dict(save_directory, config, state_dict, ckpt_format: str = "legacy"):	CODE
LOW	mcore_adapter/src/mcore_adapter/checkpointing.py	313	def generate_model_state_dict(model, ckpt_format: str = "legacy"):	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/rocm.py	53	def update_env_vars_for_visible_devices(cls, env_vars: dict, gpu_ranks: list) -> None:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/rocm.py	93	def get_vllm_run_time_env_vars(cls, gpu_rank: str) -> dict:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/cpu.py	37	def get_vllm_run_time_env_vars(cls, gpu_rank: str) -> dict:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/npu.py	67	def get_vllm_run_time_env_vars(cls, gpu_rank: str) -> dict:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/platform.py	126	def update_env_vars_for_visible_devices(cls, env_vars: dict, gpu_ranks: list) -> None:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/platform.py	162	def get_vllm_run_time_env_vars(cls, gpu_rank: str) -> dict:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/cuda.py	68	def get_vllm_run_time_env_vars(cls, gpu_rank: str) -> dict:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/unknown.py	68	def get_vllm_run_time_env_vars(cls, gpu_rank: str) -> dict:	CODE
LOW⚡	…src/mcore_adapter/parallel_functions/vocab_parallel.py	27	def vocab_range_from_per_partition_vocab_size(	CODE
LOW⚡	…src/mcore_adapter/parallel_functions/vocab_parallel.py	35	def vocab_range_from_global_vocab_size(global_vocab_size: int, rank: int, world_size: int) -> Sequence[int]:	CODE
LOW⚡	…src/mcore_adapter/parallel_functions/vocab_parallel.py	42	def calculate_predicted_logits(	CODE
LOW	…src/mcore_adapter/parallel_functions/vocab_parallel.py	188	def vocab_parallel_target_rank(vocab_parallel_logits: "torch.Tensor", target: "torch.Tensor") -> "torch.Tensor":	CODE
LOW	…dapter/parallel_functions/encoder_sequence_parallel.py	12	def _gather_and_reorder_along_first_dim(	CODE
LOW	…dapter/parallel_functions/encoder_sequence_parallel.py	129	def encoder_sequence_parallel_gather(input_, split_plan, input_size_list):	CODE
LOW	…dapter/parallel_functions/encoder_sequence_parallel.py	133	def encoder_small_batch_size_gather(input_):	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_utils.py	125	def _mca_lora_logits_postprocess(logits: "torch.Tensor"):	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_utils.py	136	def mca_lora_logits_postprocess_hook(module, input, output):	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_utils.py	149	def check_and_get_attention_backend_by_env(attention_backend: AttnBackend):	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_utils.py	171	def get_thd_data_on_this_cp_rank(	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_utils.py	197	def configure_resized_vocab_size(	CODE
LOW⚡	…ter/src/mcore_adapter/models/sequence_packing_mixin.py	86	def build_attention_mask_from_packed_seq_params(	CODE
LOW	…ter/src/mcore_adapter/models/sequence_packing_mixin.py	208	def build_multimodal_embeddings(	CODE
LOW	…ter/src/mcore_adapter/models/sequence_packing_mixin.py	286	def construct_multimodal_inputs(	CODE
LOW	…ter/src/mcore_adapter/models/sequence_packing_mixin.py	472	def _compute_full_sequence_embedding(	CODE
LOW	…ter/src/mcore_adapter/models/sequence_packing_mixin.py	518	def _pack_deepstack_visual_embeds(	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_factory.py	199	def get_batch_on_this_cp_rank(self, args, *kwargs):	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_factory.py	288	def get_batch_on_this_cp_rank(self, batch: Dict[str, "torch.Tensor"], dim3_keys: List[str] = ["attention_mask"]):	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_factory.py	325	def enable_input_require_grads(self):	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_factory.py	331	def make_inputs_require_grads(module, input, output):	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_factory.py	377	def _get_transformer_layer_spec(self, config: Optional["McaModelConfig"] = None):	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_factory.py	414	def compute_language_model_loss(self, labels: torch.Tensor, logits: torch.Tensor) -> torch.Tensor:	CODE
LOW	…mcore_adapter/models/qwen3_next/modeling_qwen3_next.py	12	def _get_transformer_layer_spec(self, config: Optional[Qwen3NextConfig] = None):	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	508	def _split_sharded_state_dict(self, sharded_state_dict: dict) -> list[dict]:	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	542	def _stream_dist_weights_from_shards(self, sharded_state_dict: dict):	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	722	def convert_checkpoint_to_mca(	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	327	def _get_num_layers_per_virtual_rank(self):	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	604	def handle_grouped_duplicated(self, name: str, weights: Union["Tensor", list["Tensor"]]) -> dict[str, "Tensor"]:	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	621	def _convert_te_grouped_column(self, name: str, weights: "Tensor"):	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	631	def _revert_te_grouped_column(self, name: str, weights: list["Tensor"], moe_index_preprocessed: bool = False):	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	869	def is_tensor_parallel_dup_weight(self, name: str) -> bool:	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	873	def is_expert_parallel_weight(self, name: str) -> bool:	CODE
LOW	…adapter/src/mcore_adapter/models/converter/template.py	473	def convert_hf_to_mca_config_kws(self, hf_config: "PretrainedConfig", **kw_args):	CODE
LOW	…er/src/mcore_adapter/models/converter/convert_utils.py	149	def remove_mca_mtp_weight_prefix(weight_name: str):	CODE
LOW	…er/src/mcore_adapter/models/converter/convert_utils.py	309	def _noisy_mean_initialization(embed_weight: "torch.Tensor", num_new_tokens: int) -> None:	CODE
LOW	…/src/mcore_adapter/models/converter/model_converter.py	90	def load_mca_state_dict_from_hf(self, model_path: str, vp_stage: int):	CODE
LOW	…/src/mcore_adapter/models/converter/model_converter.py	165	def _mca_named_params_with_vp_stage(self, models, params_group=None):	CODE
LOW	…/src/mcore_adapter/models/converter/model_converter.py	238	def _convert_moe_weight_with_allgather(	CODE
422 more matches not shown…

Unused Imports420 hits · 413 pts

Severity	File	Line	Context
LOW	mcore_adapter/src/mcore_adapter/initialize.py	9	CODE
LOW	mcore_adapter/src/mcore_adapter/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/__init__.py	2	CODE
LOW	mcore_adapter/src/mcore_adapter/__init__.py	3	CODE
LOW	mcore_adapter/src/mcore_adapter/__init__.py	3	CODE
LOW	mcore_adapter/src/mcore_adapter/checkpointing.py	7	CODE
LOW	…apter/src/mcore_adapter/parallel_functions/__init__.py	1	CODE
LOW	…apter/src/mcore_adapter/parallel_functions/__init__.py	2	CODE
LOW	…apter/src/mcore_adapter/parallel_functions/__init__.py	2	CODE
LOW	…apter/src/mcore_adapter/parallel_functions/__init__.py	3	CODE
LOW	…src/mcore_adapter/parallel_functions/fused_logprobs.py	16	CODE
LOW	…src/mcore_adapter/parallel_functions/fused_logprobs.py	18	CODE
LOW	…src/mcore_adapter/parallel_functions/fused_logprobs.py	18	CODE
LOW	…src/mcore_adapter/parallel_functions/fused_logprobs.py	18	CODE
LOW	…src/mcore_adapter/parallel_functions/fused_logprobs.py	18	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_utils.py	16	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_utils.py	175	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_config.py	20	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_config.py	27	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	1	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	21	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	21	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	22	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	23	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	23	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_factory.py	41	CODE
LOW	…dapter/src/mcore_adapter/models/qwen3_next/__init__.py	25	CODE
LOW	…dapter/src/mcore_adapter/models/qwen3_next/__init__.py	26	CODE
LOW	…e_adapter/src/mcore_adapter/models/mixtral/__init__.py	3	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	7	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	32	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	45	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	46	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	47	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	48	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	25	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	27	CODE
LOW	…adapter/src/mcore_adapter/models/converter/template.py	25	CODE
LOW	…adapter/src/mcore_adapter/models/converter/template.py	26	CODE
LOW	…/src/mcore_adapter/models/converter/model_converter.py	43	CODE
LOW	…/src/mcore_adapter/models/converter/model_converter.py	45	CODE
LOW	mcore_adapter/src/mcore_adapter/models/auto/__init__.py	1	CODE
360 more matches not shown…

Modern AI Meta-Vocabulary87 hits · 276 pts

Severity	File	Line	Snippet	Context
MEDIUM	README.md	49	\| [01/01/2026] 🎉 Our [Let It Flow: Agentic Crafting on Rock and Roll](https://arxiv.org/abs/2512.24873) report relea	CODE
MEDIUM⚡	README.md	160	- [Freshness-Aware-PER](https://arxiv.org/abs/2604.16918): A freshness-aware prioritized experience replay framework for	CODE
MEDIUM⚡	README.md	165	- [STAgent](https://arxiv.org/abs/2512.24957): An agentic LLM specialized for spatio-temporal understanding and complex	CODE
MEDIUM⚡	README.md	168	- [EARL](https://arxiv.org/abs/2510.05943): Efficient Agentic RL Systems for LLMs, introducing a dynamic parallelism sel	CODE
MEDIUM	roll/pipeline/agentic/agentic_config.py	187	# agentic related	COMMENT
MEDIUM	roll/distributed/scheduler/router.py	233	# In agentic pipeline (pure DP): rank == dp_rank, so directly access rank2devices	COMMENT
MEDIUM⚡	tests/pipeline/agentic_pipeline_config.yaml	24	# project: roll-agentic	COMMENT
MEDIUM⚡	tests/pipeline/agentic_pipeline_config.yaml	28	# - agentic	COMMENT
MEDIUM⚡	tests/pipeline/agentic_pipeline_config.yaml	35	# project: roll-agentic	COMMENT
MEDIUM⚡	tests/pipeline/agentic_pipeline_config.yaml	37	# experiment_name: roll-agentic-examples	COMMENT
MEDIUM⚡	tests/pipeline/agentic_pipeline_config.yaml	40	# - agentic	COMMENT
MEDIUM⚡	…2.5-7B-agentic_megatron/agentic_val_webshop_gigpo.yaml	20	# project: roll-agentic	COMMENT
MEDIUM⚡	…2.5-7B-agentic_megatron/agentic_val_webshop_gigpo.yaml	24	# - agentic	COMMENT
MEDIUM⚡	…2.5-7B-agentic_megatron/agentic_val_webshop_gigpo.yaml	32	# project: roll-agentic	COMMENT
MEDIUM⚡	…2.5-7B-agentic_megatron/agentic_val_webshop_gigpo.yaml	37	# - agentic	COMMENT
MEDIUM	…en2.5-7B-agentic_megatron/agentic_val_webshop_amd.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…en2.5-7B-agentic_megatron/agentic_val_webshop_amd.yaml	24	# - agentic	COMMENT
MEDIUM⚡	…s/qwen2.5-7B-agentic_megatron/agentic_val_webshop.yaml	20	# project: roll-agentic	COMMENT
MEDIUM⚡	…s/qwen2.5-7B-agentic_megatron/agentic_val_webshop.yaml	24	# - agentic	COMMENT
MEDIUM⚡	…s/qwen2.5-7B-agentic_megatron/agentic_val_webshop.yaml	32	# project: roll-agentic	COMMENT
MEDIUM⚡	…s/qwen2.5-7B-agentic_megatron/agentic_val_webshop.yaml	37	# - agentic	COMMENT
MEDIUM	…en3-30BA3B-agentic_fsdp2/agentic_val_sokoban_30a3.yaml	25	# project: roll-agentic	COMMENT
MEDIUM	…en3-30BA3B-agentic_fsdp2/agentic_val_sokoban_30a3.yaml	29	# - agentic	COMMENT
MEDIUM	examples/qwen2.5-vl-3B-agentic/agentic_val_sokoban.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	examples/qwen2.5-vl-3B-agentic/agentic_val_sokoban.yaml	24	# - agentic	COMMENT
MEDIUM	…es/qwen2.5-0.5B-agentic/agentic_val_sokoban_gigpo.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…es/qwen2.5-0.5B-agentic/agentic_val_sokoban_gigpo.yaml	24	# - agentic	COMMENT
MEDIUM	…B-agentic/agent_val_frozen_lake-pg_var_is_correct.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…B-agentic/agent_val_frozen_lake-pg_var_is_correct.yaml	24	# - agentic	COMMENT
MEDIUM	…0.5B-agentic/agentic_val_sokoban_dynamic_batching.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…0.5B-agentic/agentic_val_sokoban_dynamic_batching.yaml	24	# - agentic	COMMENT
MEDIUM	…/qwen2.5-0.5B-agentic/agent_val_frozen_lake_gigpo.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…/qwen2.5-0.5B-agentic/agent_val_frozen_lake_gigpo.yaml	24	# - agentic	COMMENT
MEDIUM	…n2.5-0.5B-agentic/agent_val_frozen_lake_async_amd.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…n2.5-0.5B-agentic/agent_val_frozen_lake_async_amd.yaml	24	# - agentic	COMMENT
MEDIUM	…/qwen2.5-0.5B-agentic/agentic_val_sokoban_sandbox.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…/qwen2.5-0.5B-agentic/agentic_val_sokoban_sandbox.yaml	24	# - agentic	COMMENT
MEDIUM	…qwen2.5-0.5B-agentic/agent_val_frozen_lake-pg_var.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…qwen2.5-0.5B-agentic/agent_val_frozen_lake-pg_var.yaml	24	# - agentic	COMMENT
MEDIUM	…les/qwen2.5-0.5B-agentic/agentic_val_sokoban_lora.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…les/qwen2.5-0.5B-agentic/agentic_val_sokoban_lora.yaml	24	# - agentic	COMMENT
MEDIUM	…s/qwen2.5-0.5B-agentic/agentic_val_sokoban_native.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…s/qwen2.5-0.5B-agentic/agentic_val_sokoban_native.yaml	24	# - agentic	COMMENT
MEDIUM	…ples/qwen2.5-0.5B-agentic/agentic_rollout_sokoban.yaml	16	# project: roll-agentic	COMMENT
MEDIUM	…ples/qwen2.5-0.5B-agentic/agentic_rollout_sokoban.yaml	20	# - agentic	COMMENT
MEDIUM	examples/qwen2.5-0.5B-agentic/agentic_val_sokoban.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	examples/qwen2.5-0.5B-agentic/agentic_val_sokoban.yaml	24	# - agentic	COMMENT
MEDIUM	…es/qwen2.5-0.5B-agentic/agent_val_frozen_lake_amd.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…es/qwen2.5-0.5B-agentic/agent_val_frozen_lake_amd.yaml	24	# - agentic	COMMENT
MEDIUM	…amples/qwen2.5-0.5B-agentic/agent_val_frozen_lake.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…amples/qwen2.5-0.5B-agentic/agent_val_frozen_lake.yaml	24	# - agentic	COMMENT
MEDIUM	…/qwen2.5-0.5B-agentic/agent_val_frozen_lake_async.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…/qwen2.5-0.5B-agentic/agent_val_frozen_lake_async.yaml	24	# - agentic	COMMENT
MEDIUM	…2.5-0.5B-agentic/agentic_val_sokoban_agent_runner.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…2.5-0.5B-agentic/agentic_val_sokoban_agent_runner.yaml	24	# - agentic	COMMENT
MEDIUM	examples/agentic_deepeyes/deepeyes.yaml	10	# pipeline_cls defaults to roll.pipeline.agentic.agentic_pipeline.AgenticPipeline	COMMENT
MEDIUM	…entic_demo/agent_val_frozen_lake_multi_nodes_demo.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	…entic_demo/agent_val_frozen_lake_multi_nodes_demo.yaml	24	# - agentic	COMMENT
MEDIUM	examples/agentic_demo/agent_val_sokoban_sandbox.yaml	20	# project: roll-agentic	COMMENT
MEDIUM	examples/agentic_demo/agent_val_sokoban_sandbox.yaml	24	# - agentic	COMMENT
27 more matches not shown…

Decorative Section Separators66 hits · 216 pts

Severity	File	Line	Snippet	Context
MEDIUM⚡	…ter/src/mcore_adapter/models/sequence_packing_mixin.py	77	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	…ter/src/mcore_adapter/models/sequence_packing_mixin.py	79	# ------------------------------------------------------------------	COMMENT
MEDIUM	…ter/src/mcore_adapter/models/sequence_packing_mixin.py	308	# ------------------------------------------------------------------	COMMENT
MEDIUM	…ter/src/mcore_adapter/models/sequence_packing_mixin.py	310	# ------------------------------------------------------------------	COMMENT
MEDIUM	…/pipeline/agentic/proxy/router/ingress_proxy_router.py	85	# ------------------------------------------------------------------	COMMENT
MEDIUM	…/pipeline/agentic/proxy/router/ingress_proxy_router.py	87	# ------------------------------------------------------------------	COMMENT
MEDIUM	…/pipeline/agentic/proxy/router/ingress_proxy_router.py	197	# ------------------------------------------------------------------	COMMENT
MEDIUM	…/pipeline/agentic/proxy/router/ingress_proxy_router.py	199	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	1186	# ---------------------------------------------------------------------------	COMMENT
MEDIUM⚡	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	1188	# ---------------------------------------------------------------------------	COMMENT
MEDIUM	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	69	# ------------------------------------------------------------------	COMMENT
MEDIUM	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	71	# ------------------------------------------------------------------	COMMENT
MEDIUM	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	85	# ------------------------------------------------------------------	COMMENT
MEDIUM	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	87	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	108	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	110	# ------------------------------------------------------------------	COMMENT
MEDIUM	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	159	# ------------------------------------------------------------------	COMMENT
MEDIUM	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	161	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	301	# ------------------------------------------------------------------	COMMENT
MEDIUM⚡	…ipeline/agentic/agent_runner/rock/rock_agent_runner.py	303	# ------------------------------------------------------------------	COMMENT
MEDIUM	roll/distributed/scheduler/remote_protocol.py	254	# ============================================================================	COMMENT
MEDIUM	roll/distributed/scheduler/remote_protocol.py	259	# ============================================================================	COMMENT
MEDIUM	roll/utils/fsdp_utils.py	474	# ============================================================================	COMMENT
MEDIUM	roll/utils/fsdp_utils.py	481	# ============================================================================	COMMENT
MEDIUM	roll/models/model_providers.py	201	# ---------------------------------------------------------------------	STRING
MEDIUM	roll/third_party/fsdp2/tiled_mlp.py	166	# ============================================================================	COMMENT
MEDIUM	roll/third_party/fsdp2/tiled_mlp.py	168	# ============================================================================	COMMENT
MEDIUM⚡	roll/third_party/vllm/gdn_patcher.py	210	# ================================================================	COMMENT
MEDIUM⚡	roll/third_party/vllm/gdn_patcher.py	213	# ================================================================	COMMENT
MEDIUM⚡	roll/third_party/vllm/gdn_patcher.py	219	# ================================================================	COMMENT
MEDIUM	roll/third_party/megatron/mtp_patcher.py	47	# ============================================================================	COMMENT
MEDIUM	roll/third_party/megatron/mtp_patcher.py	49	# ============================================================================	COMMENT
MEDIUM⚡	roll/third_party/megatron/mtp_patcher.py	222	# ============================================================================	COMMENT
MEDIUM⚡	roll/third_party/megatron/mtp_patcher.py	224	# ============================================================================	COMMENT
MEDIUM	roll/third_party/megatron/mtp_patcher.py	285	# ============================================================================	COMMENT
MEDIUM	roll/third_party/megatron/mtp_patcher.py	287	# ============================================================================	COMMENT
MEDIUM	roll/third_party/megatron/util.py	65	# ----------------------------------------------------------------------------	COMMENT
MEDIUM	roll/third_party/megatron/util.py	68	# ----------------------------------------------------------------------------	COMMENT
MEDIUM	roll/third_party/megatron/util.py	136	# -------------------------------------------------------------------------	COMMENT
MEDIUM	roll/third_party/megatron/util.py	139	# -------------------------------------------------------------------------	COMMENT
MEDIUM	…ted/strategy/log_probs/test_fsdp_vlm_log_probs_perf.py	72	# ------------------------------------------------------------------	COMMENT
MEDIUM	…ts/distributed/strategy/log_probs/apply_model_patch.py	512	# -----------------------------	COMMENT
MEDIUM	…ts/distributed/strategy/log_probs/apply_model_patch.py	514	# -----------------------------	COMMENT
MEDIUM	…ts/distributed/strategy/log_probs/apply_model_patch.py	526	# -------------------------	COMMENT
MEDIUM	…ts/distributed/strategy/log_probs/apply_model_patch.py	529	# -------------------------	COMMENT
MEDIUM	…ts/distributed/strategy/log_probs/apply_model_patch.py	577	# -------------------------	COMMENT
MEDIUM	…ts/distributed/strategy/log_probs/apply_model_patch.py	581	# -------------------------	COMMENT
MEDIUM	…tributed/strategy/log_probs/test_fsdp_vlm_log_probs.py	43	# ------------------------------------------------------------------	COMMENT
MEDIUM	examples/docs_examples/example_grpo.yaml	38	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_grpo.yaml	73	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_ppo.yaml	38	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_ppo.yaml	44	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_ppo.yaml	77	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_raft_pp.yaml	38	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_raft_pp.yaml	73	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_raft_pp.yaml	91	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_reinforce_pp.yaml	38	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_reinforce_pp.yaml	73	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_reinforce_pp.yaml	91	# --------------------------	COMMENT
MEDIUM	examples/docs_examples/example_gspo.yaml	38	# --------------------------	COMMENT
6 more matches not shown…

Excessive Try-Catch Wrapping188 hits · 215 pts

Severity	File	Line	Snippet	Context
LOW	mcore_adapter/src/mcore_adapter/training_args.py	281	except Exception as e:	CODE
LOW	mcore_adapter/src/mcore_adapter/utils.py	76	except Exception:	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	159	except Exception as e:	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	784	except Exception as e:	CODE
LOW	roll/pipeline/base_pipeline.py	183	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/rlvr_pipeline.py	446	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/actor_worker.py	187	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/rewards/video_r1_reward_worker.py	44	except Exception as e:	CODE
MEDIUM	roll/pipeline/rlvr/rewards/video_r1_reward_worker.py	45	print(f"Error converting '{num_str}' to float: {e}")	CODE
MEDIUM	roll/pipeline/rlvr/rewards/video_r1_reward_worker.py	40	def normalize_number(num_str):	CODE
LOW⚡	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1160	except Exception:	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1279	except Exception:	CODE
LOW	…vr/rewards/multiple_choice_boxed_rule_reward_worker.py	143	except Exception as e:	CODE
LOW	…l/pipeline/rlvr/rewards/remote_reward_system_worker.py	165	except Exception as e:	CODE
LOW	…l/pipeline/rlvr/rewards/remote_reward_system_worker.py	211	except Exception as e:	CODE
LOW	…l/pipeline/rlvr/rewards/remote_reward_system_worker.py	258	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/rewards/ifeval_rule_reward_worker.py	629	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/rewards/math_rule_reward_worker.py	179	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/rewards/math_rule_reward_worker.py	197	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/rewards/math_rule_reward_worker.py	292	except Exception as e:	CODE
MEDIUM	roll/pipeline/rlvr/rewards/math_rule_reward_worker.py	126	def _hf_verify_math_sample(response, answer, result, prompt):	CODE
LOW	…peline/rlvr/rewards/crossthinkqa_rule_reward_worker.py	230	except Exception as e:	CODE
LOW	…ipeline/rlvr/rewards/general_val_rule_reward_worker.py	164	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/rewards/llm_judge_reward_worker.py	99	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/rewards/llm_judge_reward_worker.py	148	except Exception as e:	CODE
LOW	roll/pipeline/rlvr/rewards/llm_judge_reward_worker.py	162	except Exception as e:	CODE
MEDIUM	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	621	print(f"Error executing test case: {e}")	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	200	except Exception as e:	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	249	except Exception as e:	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	288	except Exception as e:	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	354	except Exception as e: # Added specific exception type	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	525	except Exception as e:	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	586	except Exception as e:	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	620	except Exception as e:	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	704	except Exception as e:	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	770	except Exception as e:	CODE
LOW	roll/pipeline/distill/distill_vlm_pipeline.py	66	except Exception as e:	CODE
LOW	roll/pipeline/agentic/utils.py	55	except Exception as e:	CODE
LOW	roll/pipeline/agentic/utils.py	277	except Exception as e:	CODE
MEDIUM	roll/pipeline/agentic/utils.py	279	print(f"Error saving gif: {e}")	CODE
LOW	roll/pipeline/agentic/environment_worker.py	124	except Exception as e:	CODE
LOW	roll/pipeline/agentic/tools/action_parser.py	62	except Exception:	CODE
LOW	roll/pipeline/agentic/tools/action_parser.py	75	except Exception:	CODE
LOW	roll/pipeline/agentic/tools/action_parser.py	102	except Exception as e:	CODE
LOW	roll/pipeline/agentic/tools/action_parser.py	131	except Exception as e:	CODE
LOW	roll/pipeline/agentic/tools/action_parser.py	149	except Exception as e:	CODE
LOW	roll/pipeline/agentic/tools/mcp_tool.py	128	except Exception as e:	CODE
LOW	roll/pipeline/agentic/tools/mcp_tool.py	148	except Exception as e:	CODE
LOW	roll/pipeline/agentic/proxy/proxy_server.py	82	except Exception as e:	CODE
LOW	roll/pipeline/agentic/proxy/proxy_server.py	131	except Exception as e:	CODE
LOW	…/pipeline/agentic/proxy/router/ingress_proxy_router.py	370	except Exception as e:	CODE
MEDIUM	…/pipeline/agentic/proxy/router/ingress_proxy_router.py	371	print(f"Error: {e}")	CODE
MEDIUM⚡	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	1182	print(f"Error: {e}")	CODE
MEDIUM	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	1131	print(f"Error occurred while listing rules: {error.message if hasattr(error, 'message') else str(error)}")	CODE
LOW⚡	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	709	except Exception as e:	CODE
LOW⚡	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	1181	except Exception as e:	CODE
LOW	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	440	except Exception as e:	CODE
LOW	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	498	except Exception as e:	CODE
LOW	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	617	except Exception as e:	CODE
LOW	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	762	except Exception as e:	CODE
128 more matches not shown…

Deep Nesting189 hits · 181 pts

Severity	File	Line	Context
LOW	mcore_adapter/src/mcore_adapter/patcher.py	85	CODE
LOW	mcore_adapter/src/mcore_adapter/patcher.py	91	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_config.py	54	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_factory.py	58	CODE
LOW	mcore_adapter/src/mcore_adapter/models/model_factory.py	92	CODE
LOW	…dapter/src/mcore_adapter/models/qwen3_next/__init__.py	98	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	96	CODE
LOW	…r/src/mcore_adapter/models/converter/post_converter.py	280	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	552	CODE
LOW	…r/src/mcore_adapter/models/converter/dist_converter.py	586	CODE
LOW	…adapter/src/mcore_adapter/models/converter/template.py	96	CODE
LOW	…adapter/src/mcore_adapter/models/converter/template.py	570	CODE
LOW	…/src/mcore_adapter/models/converter/model_converter.py	138	CODE
LOW	…/src/mcore_adapter/models/converter/model_converter.py	165	CODE
LOW	…mcore_adapter/models/qwen2_5_vl/modeling_qwen2_5_vl.py	178	CODE
LOW	…apter/src/mcore_adapter/models/qwen3_5_moe/__init__.py	83	CODE
LOW	…src/mcore_adapter/models/qwen2_vl/modeling_qwen2_vl.py	177	CODE
LOW	…e_adapter/src/mcore_adapter/models/qwen3_5/__init__.py	182	CODE
LOW	…src/mcore_adapter/models/qwen3_vl/transformer_block.py	38	CODE
LOW	…src/mcore_adapter/models/qwen3_vl/transformer_block.py	175	CODE
LOW	…dapter/src/mcore_adapter/models/qwen3_vl/rope_utils.py	182	CODE
LOW	mcore_adapter/src/mcore_adapter/adapters/utils.py	20	CODE
LOW	mcore_adapter/src/mcore_adapter/adapters/lora_layer.py	531	CODE
LOW	mcore_adapter/src/mcore_adapter/adapters/lora_layer.py	143	CODE
LOW	mcore_adapter/src/mcore_adapter/adapters/lora_layer.py	207	CODE
LOW	mcore_adapter/src/mcore_adapter/adapters/lora_layer.py	262	CODE
LOW	mcore_adapter/src/mcore_adapter/adapters/lora_layer.py	352	CODE
LOW	mcore_adapter/src/mcore_adapter/trainer/trainer.py	409	CODE
LOW	mcore_adapter/src/mcore_adapter/trainer/trainer.py	627	CODE
LOW	mcore_adapter/src/mcore_adapter/trainer/trainer.py	808	CODE
LOW	mcore_adapter/src/mcore_adapter/trainer/trainer.py	949	CODE
LOW	mcore_adapter/src/mcore_adapter/trainer/dpo_trainer.py	263	CODE
LOW	roll/pipeline/base_worker.py	63	CODE
LOW	roll/pipeline/base_pipeline.py	139	CODE
LOW	roll/pipeline/base_pipeline.py	186	CODE
LOW	roll/pipeline/rlvr/rlvr_pipeline.py	459	CODE
LOW	roll/pipeline/rlvr/actor_pg_worker.py	26	CODE
LOW	roll/pipeline/rlvr/actor_pg_worker.py	444	CODE
LOW	roll/pipeline/rlvr/rlvr_config.py	157	CODE
LOW	roll/pipeline/rlvr/rlvr_vlm_pipeline.py	329	CODE
LOW	roll/pipeline/rlvr/actor_worker.py	169	CODE
LOW	roll/pipeline/rlvr/rewards/video_r1_reward_worker.py	32	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1031	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1112	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	214	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	301	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	673	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	753	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1199	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1640	CODE
LOW	…l/pipeline/rlvr/rewards/remote_reward_system_worker.py	185	CODE
LOW	roll/pipeline/rlvr/rewards/ifeval_rule_reward_worker.py	44	CODE
LOW	roll/pipeline/rlvr/rewards/math_rule_reward_worker.py	40	CODE
LOW	roll/pipeline/rlvr/rewards/math_rule_reward_worker.py	184	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	37	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	479	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	536	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	656	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	143	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	206	CODE
129 more matches not shown…

Docstring Block Structure32 hits · 160 pts

Severity	File	Line	Snippet	Context
HIGH	mcore_adapter/src/mcore_adapter/platforms/platform.py	163	Generate the runtime environment variables required for vLLM execution on the specified GPU rank.	STRING
HIGH	…src/mcore_adapter/parallel_functions/fused_logprobs.py	261	Fused log_probs forward pass with in-place output tensors. Args: x: Input logits tensor of shape (M, N	STRING
HIGH	…src/mcore_adapter/parallel_functions/fused_logprobs.py	578	Entropy backward pass using fused kernel. This function computes the gradient of entropy with respect to the i	STRING
HIGH	…l/pipeline/rlvr/rewards/remote_reward_system_worker.py	410	Normalize various input data formats into a uniform reward reference format. Args: data (D	STRING
HIGH	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	44	Select an ALB load balancer ID from a comma-separated list based on hash of job_name and listen_port. This fun	STRING
HIGH	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	98	Discover all ALBs in the current region and set them to environment variable ALB_LOAD_BALANCER_IDS. This funct	STRING
HIGH	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	400	Poll listener status until it reaches target state. Args: client: ALB client instance listener	STRING
HIGH	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	459	Poll server group status until it reaches target state. Args: client: ALB client instance serv	STRING
HIGH	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	803	Save server group ID or listener ID to a file in ALB_RESOURCE_CLEAN_DIR. Args: resource_id: Resource I	STRING
HIGH	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	991	Poll rule status until it reaches target state. Args: client: ALB client instance rule_id: Rul	STRING
HIGH	roll/pipeline/agentic/llm_proxy/proxy_utils.py	28	Generate text through proxy with support for multimodal inputs. This function formats messages using chat temp	STRING
HIGH	roll/pipeline/agentic/env_manager/proxy_env_manager.py	184	Core inference handler shared by Push (HTTP) and Pull (ModelService) modes. Args: body: Pa	STRING
HIGH	roll/platforms/platform.py	167	Generate the runtime environment variables required for vLLM execution on the specified GPU rank.	STRING
HIGH	roll/distributed/scheduler/rollout_mock_mixin.py	162	Load pre-recorded DataProto from disk (strict mode). Args: global_step: Current training s	STRING
HIGH	roll/distributed/scheduler/protocol.py	594	Slice the DataProto and return a new DataProto object. Args: start (int, optional): Start	STRING
HIGH	roll/distributed/scheduler/rollout_scheduler.py	703	Thin wrapper: Delegate shrink operation to RequestScheduler. v4.6 ARCHITECTURAL CHANGE: RolloutScheduler no lon	STRING
HIGH	roll/distributed/scheduler/rollout_scheduler.py	744	Thin wrapper: Delegate expand operation to RequestScheduler. v4.6 ARCHITECTURAL CHANGE: RolloutScheduler no lon	STRING
HIGH	roll/distributed/scheduler/router.py	321	Complete atomic shrink operation: validate → rebalance → offload → update routing. Orchestrates the full worker	STRING
HIGH	roll/distributed/scheduler/router.py	381	Complete atomic expand operation: validate → load → rebalance → update routing. Orchestrates the full worker ex	STRING
HIGH	roll/distributed/scheduler/router.py	1031	Abort requests on shrinking workers, clear mappings for natural re-dispatch. Args: shrink_dp_ranks:	STRING
HIGH	roll/distributed/scheduler/router.py	1066	Internal implementation of shrink rebalancing. PRE-CONDITION: routing_lock MUST be held by caller (shrink_worke	STRING
HIGH	roll/distributed/scheduler/router.py	1137	Add workers and rebalance via src_rank-level abort. Args: expand_dp_ranks: DP ranks to add to activ	STRING
HIGH	roll/distributed/scheduler/router.py	1170	Internal implementation of expand rebalancing. PRE-CONDITION: routing_lock MUST be held by caller (expand_worke	STRING
HIGH	roll/distributed/strategy/strategy.py	240	Compute divergence loss between student and teacher distributions with support for distributed training.	STRING
HIGH	roll/utils/sequence_packing.py	129	Split mini_batch into micro batches with sequence packing strategy. This function: 1. Calculat	STRING
HIGH	roll/utils/hash_utils.py	7	Compute the SHA256 hash of a Python object. This function handles basic types, None, lists, tuples, dictionarie	STRING
HIGH	roll/utils/qwen_vl_utils.py	154	calculate the number of frames for video used for model inputs. Args: ele (dict): a dict contains the confi	STRING
HIGH	roll/utils/qwen_vl_utils.py	244	Calculate the start and end frame indices based on the given time range. Args: ele (dict): A dictionar	STRING
HIGH	roll/utils/config_utils.py	16	Calculate the data parallel size for Megatron. Megatron uses: DP = Total_GPUs / (TP × PP × CP × EP)	STRING
HIGH	roll/third_party/megatron/fused_entropy.py	450	Fused entropy forward pass with in-place output tensors. This function computes the entropy of each row in the	STRING
HIGH	roll/third_party/megatron/fused_entropy.py	781	Entropy backward pass using fused kernel. This function computes the gradient of entropy with respect to the i	STRING
HIGH	roll/third_party/megatron/router_replay_utils.py	169	Reorder and merge per-VPP layer blocks into a contiguous layer dimension. Given a tensor shaped as [bs*vpp_siz	STRING

Cross-File Repetition30 hits · 150 pts

Severity	File	Snippet	Context
HIGH	mcore_adapter/src/mcore_adapter/platforms/rocm.py	update environment variables to control device visibility. args: env_vars (dict): dictionary of current environment vari	STRING
HIGH	mcore_adapter/src/mcore_adapter/platforms/platform.py	update environment variables to control device visibility. args: env_vars (dict): dictionary of current environment vari	STRING
HIGH	roll/platforms/rocm.py	update environment variables to control device visibility. args: env_vars (dict): dictionary of current environment vari	STRING
HIGH	roll/platforms/platform.py	update environment variables to control device visibility. args: env_vars (dict): dictionary of current environment vari	STRING
HIGH	…mcore_adapter/models/qwen2_5_vl/modeling_qwen2_5_vl.py	inputs_embeds: [s, b, h] or [s/tp, b, h] when sequence parallel ranges: sequence range	STRING
HIGH	…mcore_adapter/models/qwen3_omni/modeling_qwen3_omni.py	inputs_embeds: [s, b, h] or [s/tp, b, h] when sequence parallel ranges: sequence range	STRING
HIGH	…src/mcore_adapter/models/qwen2_vl/modeling_qwen2_vl.py	inputs_embeds: [s, b, h] or [s/tp, b, h] when sequence parallel ranges: sequence range	STRING
HIGH	…r/src/mcore_adapter/models/qwen3_5/modeling_qwen3_5.py	inputs_embeds: [s, b, h] or [s/tp, b, h] when sequence parallel ranges: sequence range	STRING
HIGH	…mcore_adapter/models/qwen2_5_vl/modeling_qwen2_5_vl.py	calculate split plan and local data according to workload, assuming workload proportional to length args: input_lengths	STRING
HIGH	…src/mcore_adapter/models/qwen2_vl/modeling_qwen2_vl.py	calculate split plan and local data according to workload, assuming workload proportional to length args: input_lengths	STRING
HIGH	…r/src/mcore_adapter/models/qwen3_5/modeling_qwen3_5.py	calculate split plan and local data according to workload, assuming workload proportional to length args: input_lengths	STRING
HIGH	…src/mcore_adapter/models/qwen3_vl/modeling_qwen3_vl.py	calculate split plan and local data according to workload, assuming workload proportional to length args: input_lengths	STRING
HIGH	roll/pipeline/base_worker.py	loss func接口定义: data: dataproto, 由train_step透传 output_tensor: torch.tensor, model.forward()的输出tensor	STRING
HIGH	roll/pipeline/rlvr/actor_pg_worker.py	loss func接口定义: data: dataproto, 由train_step透传 output_tensor: torch.tensor, model.forward()的输出tensor	STRING
HIGH	roll/pipeline/rlvr/actor_worker.py	loss func接口定义: data: dataproto, 由train_step透传 output_tensor: torch.tensor, model.forward()的输出tensor	STRING
HIGH	roll/pipeline/agentic/agentic_actor_pg_worker.py	loss func接口定义: data: dataproto, 由train_step透传 output_tensor: torch.tensor, model.forward()的输出tensor	STRING
HIGH	roll/pipeline/agentic/agentic_actor_worker.py	loss func接口定义: data: dataproto, 由train_step透传 output_tensor: torch.tensor, model.forward()的输出tensor	STRING
HIGH	roll/pipeline/dpo/actor_worker.py	loss func接口定义: data: dataproto, 由train_step透传 output_tensor: torch.tensor, model.forward()的输出tensor	STRING
HIGH	…vr/rewards/multiple_choice_boxed_rule_reward_worker.py	提取输入字符串中最后一个"end_think"标签之后的内容，并移除结果字符串开头的所有换行符。 args: input_string: 原始字符串。 returns: 提取并处理后的字符串。如果未找到"end_think"标签，则返回空	STRING
HIGH	…peline/rlvr/rewards/crossthinkqa_rule_reward_worker.py	提取输入字符串中最后一个"end_think"标签之后的内容，并移除结果字符串开头的所有换行符。 args: input_string: 原始字符串。 returns: 提取并处理后的字符串。如果未找到"end_think"标签，则返回空	STRING
HIGH	…ipeline/rlvr/rewards/general_val_rule_reward_worker.py	提取输入字符串中最后一个"end_think"标签之后的内容，并移除结果字符串开头的所有换行符。 args: input_string: 原始字符串。 returns: 提取并处理后的字符串。如果未找到"end_think"标签，则返回空	STRING
HIGH	roll/pipeline/rlvr/rewards/ifeval_rule_reward_worker.py	一个示例 reward worker，用于执行 ifeval 验证并把每个 func 的结果放到 output.tensors 中。在此示例里，ground_truths的str	STRING
HIGH	…peline/rlvr/rewards/crossthinkqa_rule_reward_worker.py	一个示例 reward worker，用于执行 ifeval 验证并把每个 func 的结果放到 output.tensors 中。在此示例里，ground_truths的str	STRING
HIGH	…ipeline/rlvr/rewards/general_val_rule_reward_worker.py	一个示例 reward worker，用于执行 ifeval 验证并把每个 func 的结果放到 output.tensors 中。在此示例里，ground_truths的str	STRING
HIGH	roll/pipeline/agentic/env_manager/base_env_manager.py	1. each time run_rollout_loop is called, it will continuously play episodes until it receives a command that data collec	STRING
HIGH	roll/pipeline/agentic/env_manager/traj_env_manager.py	1. each time run_rollout_loop is called, it will continuously play episodes until it receives a command that data collec	STRING
HIGH	…ll/pipeline/agentic/env_manager/vl_traj_env_manager.py	1. each time run_rollout_loop is called, it will continuously play episodes until it receives a command that data collec	STRING
HIGH	roll/distributed/strategy/megatron_strategy.py	input_ids [[p, p, r, r, r, 0, 0]] p: prompt, r: response, 0: pad response_mask [[0, 0, 1, 1, 1, 0, 0]]	STRING
HIGH	roll/distributed/strategy/fsdp2_strategy.py	input_ids [[p, p, r, r, r, 0, 0]] p: prompt, r: response, 0: pad response_mask [[0, 0, 1, 1, 1, 0, 0]]	STRING
HIGH	roll/models/func_providers.py	input_ids [[p, p, r, r, r, 0, 0]] p: prompt, r: response, 0: pad response_mask [[0, 0, 1, 1, 1, 0, 0]]	STRING

Structural Annotation Overuse75 hits · 130 pts

Severity	File	Line	Snippet	Context
LOW	roll/pipeline/agentic/agentic_pipeline.py	268	# Step 0: active_dp_ranks initialized with all ranks {0,1,2,3}, no expand needed	COMMENT
LOW⚡	roll/pipeline/agentic/tools/mcp_tool.py	339	# Step 1: Find the schema for the requested tool.	STRING
LOW⚡	roll/pipeline/agentic/tools/mcp_tool.py	347	# Step 2: Use the jsonschema library to validate the parameters.	STRING
LOW⚡	roll/utils/sequence_packing.py	206	# Step 4: Round up to be divisible by vp_size	COMMENT
LOW⚡	roll/utils/sequence_packing.py	210	# Step 5: Calculate workload for load balancing	COMMENT
LOW⚡	roll/utils/sequence_packing.py	215	# Step 6: Partition samples into micro batches with balanced workload	COMMENT
LOW⚡	roll/utils/sequence_packing.py	222	# Step 7: Sort and reorder for better pipeline scheduling	COMMENT
LOW	roll/utils/sequence_packing.py	179	# Step 2: Calculate initial number of micro batches	COMMENT
LOW	roll/utils/sequence_packing.py	192	# Step 3: Synchronize across DP ranks (all ranks must have same count)	COMMENT
LOW	roll/utils/sequence_packing.py	245	# Step 8: Generate micro batches	COMMENT
LOW	roll/configs/base_config.py	690	# Step 1: Normalize teacher to _reference_configs dict (always Dict[str, WorkerConfig])	COMMENT
LOW	roll/configs/base_config.py	711	# Step 2: Pure OPD mode	COMMENT
LOW	roll/configs/base_config.py	732	# Step 3: Mixed OPD mode	COMMENT
LOW	…s_roll/docs/User Guides/Pipeline/dpo_pipeline_start.md	188	### Step 1: Configure Settings	COMMENT
LOW	…s_roll/docs/User Guides/Pipeline/dpo_pipeline_start.md	198	### Step 2: Prepare Environment and Dependencies	COMMENT
LOW⚡	…s_roll/docs/User Guides/Pipeline/dpo_pipeline_start.md	210	### Step 3: Launch the Pipeline	COMMENT
LOW⚡	…s_roll/docs/User Guides/Pipeline/dpo_pipeline_start.md	218	### Step 4: Monitoring	COMMENT
LOW⚡	…s_roll/docs/User Guides/Pipeline/dpo_pipeline_start.md	228	### Step 5: Outputs and Results	COMMENT
LOW	…ll/docs/User Guides/Pipeline/distill_pipeline_start.md	193	### Step 1: Configure Settings	COMMENT
LOW	…ll/docs/User Guides/Pipeline/distill_pipeline_start.md	203	### Step 2: Prepare Environment and Dependencies	COMMENT
LOW⚡	…ll/docs/User Guides/Pipeline/distill_pipeline_start.md	215	### Step 3: Launch the Pipeline	COMMENT
LOW⚡	…ll/docs/User Guides/Pipeline/distill_pipeline_start.md	223	### Step 4: Monitoring	COMMENT
LOW⚡	…ll/docs/User Guides/Pipeline/distill_pipeline_start.md	233	### Step 5: Outputs and Results	COMMENT
LOW	…ll/docs/User Guides/Pipeline/vl_rlvr_pipeline_start.md	287	### Step 1: Configure Settings	COMMENT
LOW	…ll/docs/User Guides/Pipeline/vl_rlvr_pipeline_start.md	298	### Step 2: Prepare Environment and Dependencies	COMMENT
LOW⚡	…ll/docs/User Guides/Pipeline/vl_rlvr_pipeline_start.md	310	### Step 3: Launch the Pipeline	COMMENT
LOW⚡	…ll/docs/User Guides/Pipeline/vl_rlvr_pipeline_start.md	318	### Step 4: Monitoring	COMMENT
LOW⚡	…ll/docs/User Guides/Pipeline/vl_rlvr_pipeline_start.md	328	### Step 5: Outputs and Results	COMMENT
LOW	…ll/docs/User Guides/Pipeline/agentic_pipeline_start.md	205	### Step 1: Configuration Setup	COMMENT
LOW	…ll/docs/User Guides/Pipeline/agentic_pipeline_start.md	216	### Step 2: Environment and Dependency Preparation	COMMENT
LOW⚡	…ll/docs/User Guides/Pipeline/agentic_pipeline_start.md	228	### Step 3: Starting the Pipeline	COMMENT
LOW⚡	…ll/docs/User Guides/Pipeline/agentic_pipeline_start.md	236	### Step 4: Monitoring	COMMENT
LOW⚡	…ll/docs/User Guides/Pipeline/agentic_pipeline_start.md	246	### Step 5: Output and Results	COMMENT
LOW	…er Guides/Pipeline/on_policy_distill_pipeline_start.md	348	### Step 1: Configuration Setup	COMMENT
LOW	…er Guides/Pipeline/on_policy_distill_pipeline_start.md	359	### Step 2: Prepare Environment and Dependencies	COMMENT
LOW⚡	…er Guides/Pipeline/on_policy_distill_pipeline_start.md	371	### Step 3: Launch the Pipeline	COMMENT
LOW⚡	…er Guides/Pipeline/on_policy_distill_pipeline_start.md	379	### Step 4: Monitoring	COMMENT
LOW⚡	…er Guides/Pipeline/on_policy_distill_pipeline_start.md	389	### Step 5: Outputs and Results	COMMENT
LOW	…s_roll/docs/User Guides/Pipeline/sft_pipeline_start.md	215	### Step 1: Configuration	COMMENT
LOW	…s_roll/docs/User Guides/Pipeline/sft_pipeline_start.md	228	### Step 2: Prepare Environment and Dependencies	COMMENT
LOW⚡	…s_roll/docs/User Guides/Pipeline/sft_pipeline_start.md	239	### Step 3: Launch the Pipeline	COMMENT
LOW⚡	…s_roll/docs/User Guides/Pipeline/sft_pipeline_start.md	247	### Step 4: Monitoring	COMMENT
LOW⚡	…s_roll/docs/User Guides/Pipeline/sft_pipeline_start.md	256	### Step 5: Outputs and Results	COMMENT
LOW	…_roll/docs/User Guides/Pipeline/rlvr_pipeline_start.md	301	### Step 1: Configure Settings	COMMENT
LOW	…_roll/docs/User Guides/Pipeline/rlvr_pipeline_start.md	312	### Step 2: Prepare Environment and Dependencies	COMMENT
LOW⚡	…_roll/docs/User Guides/Pipeline/rlvr_pipeline_start.md	324	### Step 3: Launch the Pipeline	COMMENT
LOW⚡	…_roll/docs/User Guides/Pipeline/rlvr_pipeline_start.md	332	### Step 4: Monitoring	COMMENT
LOW⚡	…_roll/docs/User Guides/Pipeline/rlvr_pipeline_start.md	342	### Step 5: Outputs and Results	COMMENT
LOW	…l/docs/User Guides/Hardware Support/ascend_npu_rlvr.md	15	## Step 1: Environment Setup	COMMENT
LOW	…l/docs/User Guides/Hardware Support/ascend_npu_rlvr.md	96	## Step 2: Data Preparation	COMMENT
LOW	…l/docs/User Guides/Hardware Support/ascend_npu_rlvr.md	185	## Step 3: Model Preparation	COMMENT
LOW	…l/docs/User Guides/Hardware Support/ascend_npu_rlvr.md	219	## Step 4: Write the NPU Configuration	COMMENT
LOW	…l/docs/User Guides/Hardware Support/ascend_npu_rlvr.md	465	## Step 5: Launch Training	COMMENT
LOW	…l/docs/User Guides/Hardware Support/ascend_npu_rlvr.md	642	## Step 6: Monitor & Evaluate	COMMENT
LOW	…l/docs/User Guides/Hardware Support/ascend_npu_rlvr.md	687	## Step 7: Resume from Checkpoint	COMMENT
LOW	…cs/User Guides/Hardware Support/ascend_npu_examples.md	32	### Step 1: Start the Container	COMMENT
LOW	…cs/User Guides/Hardware Support/ascend_npu_examples.md	62	### Step 2: Set Environment Variables	COMMENT
LOW	…cs/User Guides/Hardware Support/ascend_npu_examples.md	99	### Step 3: Create NPU Configuration File	COMMENT
LOW	…cs/User Guides/Hardware Support/ascend_npu_examples.md	256	### Step 4: Launch	COMMENT
LOW	…cs/User Guides/Hardware Support/ascend_npu_examples.md	463	### Step 1: Start Containers on All Nodes	COMMENT
15 more matches not shown…

Hallucination Indicators12 hits · 120 pts

Severity	File	Line	Snippet	Context
CRITICAL	roll/pipeline/rlvr/rlvr_pipeline.py	162	for domain in self.pipeline_config.actor_train.data_args.domain_interleave_probs.keys():	CODE
CRITICAL	roll/pipeline/rlvr/rlvr_pipeline.py	633	self.pipeline_config.actor_train.strategy_args.strategy_config.get("pipeline_model_paral	CODE
CRITICAL	roll/pipeline/rlvr/rlvr_pipeline.py	634	self.pipeline_config.actor_train.strategy_args.strategy_config.get("virtual_pipeline_mod	CODE
CRITICAL	roll/pipeline/rlvr/rlvr_pipeline.py	788	self.pipeline_config.actor_train.strategy_args.strategy_config.get("pipeline_model_p	CODE
CRITICAL	roll/pipeline/rlvr/rlvr_pipeline.py	789	self.pipeline_config.actor_train.strategy_args.strategy_config.get("virtual_pipeline	CODE
CRITICAL	roll/pipeline/rlvr/rlvr_rollout_pipeline.py	35	assert self.pipeline_config.actor_infer.strategy_args.strategy_config.get("load_format", "dummy") != "dummy"	CODE
CRITICAL	roll/pipeline/rlvr/rlvr_rollout_pipeline.py	39	assert self.pipeline_config.actor_infer.strategy_args.strategy_config.get("sleep_level", 1) == 1, (	CODE
CRITICAL	roll/pipeline/rlvr/rlvr_vlm_pipeline.py	152	for domain in self.pipeline_config.actor_train.data_args.domain_interleave_probs.keys():	CODE
CRITICAL	roll/pipeline/agentic/agentic_pipeline.py	428	self.pipeline_config.actor_train.strategy_args.strategy_config.get("pipeline_model_p	CODE
CRITICAL	roll/pipeline/agentic/agentic_pipeline.py	429	self.pipeline_config.actor_train.strategy_args.strategy_config.get("virtual_pipeline	CODE
CRITICAL	roll/pipeline/agentic/agentic_pipeline.py	537	self.pipeline_config.actor_train.strategy_args.strategy_config.get("pipeline_model_p	CODE
CRITICAL	roll/pipeline/agentic/agentic_pipeline.py	538	self.pipeline_config.actor_train.strategy_args.strategy_config.get("virtual_pipeline	CODE

Self-Referential Comments25 hits · 75 pts

Severity	File	Line	Snippet	Context
MEDIUM	…src/mcore_adapter/parallel_functions/vocab_parallel.py	56	# Create a mask of valid vocab ids (1 means it needs to be masked).	COMMENT
MEDIUM	…src/mcore_adapter/parallel_functions/vocab_parallel.py	210	# Create a mask of valid vocab ids (1 means it needs to be masked).	COMMENT
MEDIUM	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1291	# Initialize COCO object for ground truth	COMMENT
MEDIUM	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1315	# Initialize COCO object for ground truth	COMMENT
MEDIUM	roll/distributed/scheduler/protocol.py	620	# Create a slice object	COMMENT
MEDIUM	roll/distributed/strategy/megatron_strategy.py	629	# Create a mock output tensor when the sample is empty to ensure the subsequent pipeline works corre	COMMENT
MEDIUM	roll/datasets/collator.py	202	# Create a batched list from single object, differ with hf's single element list	COMMENT
MEDIUM	roll/utils/local_code/testing_util.py	2	# This file is a slightly modified version of the original code from LiveCodeBench:	COMMENT
MEDIUM	roll/utils/context_parallel/ulysses_attention.py	1	# This file is modified from https://github.com/feifeibear/long-context-attention	COMMENT
MEDIUM	roll/utils/context_parallel/globals.py	1	# This file is modified from https://github.com/feifeibear/long-context-attention	COMMENT
MEDIUM	roll/utils/context_parallel/all_to_all.py	1	# This file is modified from https://github.com/feifeibear/long-context-attention	COMMENT
MEDIUM	roll/utils/context_parallel/hf_flash_attention_patch.py	264	# Create a dummy that matches the query batch size.	COMMENT
MEDIUM	roll/third_party/vllm/ray_distributed_executor.py	56	# Create the parallel GPU workers.	COMMENT
MEDIUM	…ird_party/vllm/vllm_0_12_0/ray_distributed_executor.py	48	# Create the parallel GPU workers.	COMMENT
MEDIUM	…ird_party/vllm/vllm_0_10_2/ray_distributed_executor.py	70	# Create the parallel GPU workers.	COMMENT
MEDIUM	…ird_party/vllm/vllm_0_10_2/ray_distributed_executor.py	104	# Create the workers.	COMMENT
MEDIUM	…ird_party/vllm/vllm_0_11_0/ray_distributed_executor.py	69	# Create the parallel GPU workers.	COMMENT
MEDIUM	…ird_party/vllm/vllm_0_11_0/ray_distributed_executor.py	103	# Create the workers.	COMMENT
MEDIUM	…hird_party/vllm/vllm_0_8_4/ray_distributed_executor.py	69	# Create the parallel GPU workers.	COMMENT
MEDIUM	…hird_party/vllm/vllm_0_8_4/ray_distributed_executor.py	103	# Create the workers.	COMMENT
MEDIUM	…distributed/strategy/test_vllm_strategy_beam_search.py	233	# Create a batch with dummy tensors to satisfy DataProto requirements	COMMENT
MEDIUM	…/distributed/strategy/grad_norm/test_grad_norm_unit.py	20	# Create a parameter with known gradient	COMMENT
MEDIUM	…/distributed/strategy/grad_norm/test_grad_norm_unit.py	75	# Create a simple loss: L = w1[0]^2 + w1[1]^2 + w2[0]^2	COMMENT
MEDIUM	…distributed/strategy/grad_norm/test_fsdp2_grad_norm.py	233	# Create a very simple model for easy verification	COMMENT
MEDIUM	…t_parallel/test_fsdp2_cp_qwen3_hf_rmpad_equivalence.py	298	# Create a small packed input to probe	COMMENT

Magic Placeholder Names8 hits · 68 pts

Severity	File	Line	Snippet	Context
HIGH	mcore_adapter/examples/data/dpo_zh_demo.json	3735	"value": "创建一个完整的程序像这样将是一个复杂且漫长的过程，涉及图形和音频处理。然而，我可以指导您完成初始步骤，并为您提供资源，帮助您构建项目的简单文本版本。\n\n要获取天气数据，您可以使用OpenWeatherMa	CODE
HIGH	tests/pipeline/rlvr_megatron_config.yaml	25	# api_key: your_api_key	COMMENT
HIGH⚡	tests/pipeline/agentic_pipeline_config.yaml	23	# api_key: your_api_key	COMMENT
HIGH⚡	tests/pipeline/agentic_pipeline_config.yaml	34	# api_key: your_api_key	COMMENT
HIGH⚡	…2.5-7B-agentic_megatron/agentic_val_webshop_gigpo.yaml	31	# api_key: your_api_key	COMMENT
HIGH⚡	…s/qwen2.5-7B-agentic_megatron/agentic_val_webshop.yaml	31	# api_key: your_api_key	COMMENT
HIGH	…/User Guides/Tracker & Metrics/trackers_and_metrics.md	61	api_key: your_api_key # Your SwanLab API key	CODE
HIGH	…/User Guides/Tracker & Metrics/trackers_and_metrics.md	61	api_key: your_api_key # 您的 SwanLab API 密钥	CODE

AI Structural Patterns59 hits · 56 pts

Severity	File	Line	Context
LOW	mcore_adapter/src/mcore_adapter/checkpointing.py	70	CODE
LOW	mcore_adapter/src/mcore_adapter/checkpointing.py	115	CODE
LOW	…/src/mcore_adapter/models/converter/model_converter.py	51	CODE
LOW	…mcore_adapter/models/qwen2_5_vl/modeling_qwen2_5_vl.py	342	CODE
LOW	…apter/src/mcore_adapter/models/deepseek_v3/__init__.py	89	CODE
LOW	…mcore_adapter/models/qwen3_omni/modeling_qwen3_omni.py	225	CODE
LOW	…_adapter/src/mcore_adapter/models/glm4_moe/__init__.py	80	CODE
LOW	…src/mcore_adapter/models/qwen2_vl/modeling_qwen2_vl.py	322	CODE
LOW	…e_adapter/src/mcore_adapter/models/qwen3_5/__init__.py	166	CODE
LOW	…r/src/mcore_adapter/models/qwen3_5/modeling_qwen3_5.py	271	CODE
LOW	…src/mcore_adapter/models/qwen3_vl/transformer_block.py	175	CODE
LOW	…src/mcore_adapter/models/qwen3_vl/modeling_qwen3_vl.py	59	CODE
LOW	…src/mcore_adapter/models/qwen3_vl/modeling_qwen3_vl.py	416	CODE
LOW	mcore_adapter/src/mcore_adapter/adapters/lora_layer.py	38	CODE
LOW	roll/pipeline/base_worker.py	127	CODE
LOW	roll/pipeline/base_worker.py	348	CODE
LOW	roll/pipeline/base_worker.py	524	CODE
LOW	roll/pipeline/base_worker.py	650	CODE
LOW	roll/pipeline/base_worker.py	712	CODE
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1717	CODE
LOW	…vr/rewards/multiple_choice_boxed_rule_reward_worker.py	160	CODE
LOW	…l/pipeline/rlvr/rewards/remote_reward_system_worker.py	528	CODE
LOW	roll/pipeline/rlvr/rewards/ifeval_rule_reward_worker.py	665	CODE
LOW	…peline/rlvr/rewards/crossthinkqa_rule_reward_worker.py	262	CODE
LOW	…ipeline/rlvr/rewards/general_val_rule_reward_worker.py	185	CODE
LOW	…ll/pipeline/rlvr/rewards/code_sandbox_reward_worker.py	864	CODE
LOW	roll/pipeline/distill/distill_worker.py	106	CODE
LOW	roll/pipeline/distill/distill_worker.py	156	CODE
LOW	roll/pipeline/distill/distill_worker.py	182	CODE
LOW	roll/pipeline/distill/distill_worker.py	512	CODE
LOW	roll/pipeline/dpo/actor_worker.py	86	CODE
LOW	roll/pipeline/sft/sft_worker.py	39	CODE
LOW	roll/pipeline/sft/sft_worker.py	51	CODE
LOW	roll/pipeline/sft/sft_worker.py	67	CODE
LOW	roll/distributed/scheduler/decorator.py	253	CODE
LOW	roll/distributed/executor/worker.py	208	CODE
LOW	roll/distributed/strategy/sglang_strategy.py	316	CODE
LOW	roll/distributed/strategy/mock_strategy.py	77	CODE
LOW	roll/distributed/strategy/vllm_strategy.py	228	CODE
LOW	roll/distributed/strategy/vllm_strategy.py	289	CODE
LOW	roll/distributed/strategy/fsdp2_strategy.py	1259	CODE
LOW	roll/distributed/strategy/hf_strategy.py	137	CODE
LOW	roll/utils/fsdp_utils.py	388	CODE
LOW	roll/utils/fsdp_utils.py	648	CODE
LOW	roll/utils/fsdp_utils.py	686	CODE
LOW	roll/utils/collective/pg_utils.py	11	CODE
LOW	roll/utils/context_parallel/ulysses_attention.py	61	CODE
LOW	roll/utils/context_parallel/ulysses_attention.py	361	CODE
LOW	roll/utils/context_parallel/ulysses_attention.py	244	CODE
LOW	roll/models/model_providers.py	381	CODE
LOW	roll/third_party/vllm/patch_transformers.py	9	CODE
LOW	roll/third_party/vllm/patch_transformers.py	85	CODE
LOW	roll/third_party/megatron/mtp_patcher.py	50	CODE
LOW	roll/third_party/megatron/router_replay_utils.py	205	CODE
LOW	…ts/distributed/strategy/log_probs/apply_model_patch.py	45	CODE
LOW	…ts/distributed/strategy/log_probs/apply_model_patch.py	174	CODE
LOW	…ts/distributed/strategy/log_probs/apply_model_patch.py	532	CODE
LOW	tests/third_party/sglang/test_abort_http.py	31	CODE
LOW	tests/third_party/sglang/test_abort_grpc.py	32	CODE

Modern Structural Boilerplate42 hits · 42 pts

Severity	File	Line	Snippet	Context
LOW	mcore_adapter/src/mcore_adapter/__init__.py	6	__all__ = ["McaModelConfig", "McaGPTModel", "TrainingArguments", "Seq2SeqTrainingArguments", "McaTrainer"]	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/rocm.py	28	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/rocm.py	53	def update_env_vars_for_visible_devices(cls, env_vars: dict, gpu_ranks: list) -> None:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/__init__.py	55	__all__ = [	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/npu.py	26	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/platform.py	111	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/platform.py	126	def update_env_vars_for_visible_devices(cls, env_vars: dict, gpu_ranks: list) -> None:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/cuda.py	30	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	mcore_adapter/src/mcore_adapter/platforms/unknown.py	24	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	…apter/src/mcore_adapter/parallel_functions/__init__.py	6	__all__ = ["context_parallel_gather", "encoder_sequence_parallel_gather", "encoder_small_batch_size_gather", "vocab_para	CODE
LOW	mcore_adapter/src/mcore_adapter/models/__init__.py	26	__all__ = ["McaModelConfig", "McaGPTModel", "AutoConfig", "AutoModel", "VirtualModels"]	CODE
LOW	…dapter/src/mcore_adapter/models/qwen3_next/__init__.py	220	__all__ = ["Qwen3NextConfig", "Qwen3NextModel"]	CODE
LOW	mcore_adapter/src/mcore_adapter/models/auto/__init__.py	5	__all__ = ["AutoConfig", "AutoModel", "register_config", "register_model"]	CODE
LOW	…dapter/src/mcore_adapter/models/qwen2_5_vl/__init__.py	83	__all__ = ["Qwen2_5_VLConfig", "Qwen2_5_VLModel"]	CODE
LOW	…apter/src/mcore_adapter/models/deepseek_v3/__init__.py	311	__all__ = ["DeepSeekV3Model"]	CODE
LOW	…dapter/src/mcore_adapter/models/qwen3_omni/__init__.py	143	__all__ = ["Qwen3OmniMoeConfig", "Qwen3OmniMoeModel"]	CODE
LOW	…_adapter/src/mcore_adapter/models/qwen2_vl/__init__.py	83	__all__ = ["Qwen2VLConfig", "Qwen2VLModel"]	CODE
LOW	…e_adapter/src/mcore_adapter/models/qwen3_5/__init__.py	312	__all__ = ["Qwen3_5Config", "Qwen3_5Model"]	CODE
LOW	…ore_adapter/src/mcore_adapter/models/llama/__init__.py	64	__all__ = ["LlamaConfig", "LlamaModel"]	CODE
LOW	…_adapter/src/mcore_adapter/models/qwen3_vl/__init__.py	92	__all__ = ["Qwen3VLConfig", "Qwen3VLModel"]	CODE
LOW	mcore_adapter/src/mcore_adapter/adapters/__init__.py	31	__all__ = [	CODE
LOW	mcore_adapter/src/mcore_adapter/trainer/__init__.py	6	__all__ = ["McaTrainer", "DPOTrainer", "DPOConfig"]	CODE
LOW	roll/pipeline/agentic/proxy/proxy_server.py	13	logger = logging.getLogger(__name__)	CODE
LOW	roll/pipeline/agentic/proxy/router/__init__.py	9	__all__ = ["ProxyRouter", "ALBProxyRouter", "IngressProxyRouter", "batch_cleanup", "create_proxy_router"]	CODE
LOW	roll/pipeline/agentic/agent_runner/__init__.py	3	__all__ = ["AgentRunner", "EpisodeResult"]	CODE
LOW	roll/pipeline/agentic/agent_runner/rock/__init__.py	5	__all__ = ["RockAgentRunner", "PushModeRunner", "PullModeRunner"]	CODE
LOW	roll/pipeline/agentic/env_manager/message_tracker.py	195	def update_tools(self, tools: List[dict]) -> None:	CODE
LOW	roll/platforms/rocm.py	28	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	roll/platforms/rocm.py	55	def update_env_vars_for_visible_devices(cls, env_vars: dict, gpu_ranks: list) -> None:	CODE
LOW	roll/platforms/__init__.py	53	__all__ = [	CODE
LOW	roll/platforms/npu.py	29	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	roll/platforms/platform.py	109	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	roll/platforms/platform.py	130	def update_env_vars_for_visible_devices(cls, env_vars: dict, gpu_ranks: list) -> None:	CODE
LOW	roll/platforms/cuda.py	28	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	roll/platforms/unknown.py	23	def set_allocator_settings(cls, env: str) -> None:	CODE
LOW	roll/distributed/scheduler/transfer_backend.py	224	__all__ = ["init_transfer_backend", "put", "get", "delete"]	CODE
LOW	roll/utils/qwen_vl_utils.py	43	logger = logging.getLogger(__name__)	CODE
LOW	roll/utils/context_parallel/__init__.py	4	__all__ = ["set_upg_manager", "get_ulysses_group", "apply_ulysses_patch", "unapply_ulysses_patch"]	CODE
LOW	roll/third_party/vllm/gdn_patcher.py	20	logger = logging.getLogger(__name__)	CODE
LOW	roll/third_party/vllm/__init__.py	148	__all__ = ["create_async_llm"]	CODE
LOW	roll/third_party/vllm/vllm_0_8_4/__init__.py	124	__all__ = []	CODE
LOW	roll/third_party/megatron/optimizer.py	14	logger = logging.getLogger(__name__)	CODE

Over-Commented Block40 hits · 40 pts

Severity	File	Line	Snippet	Context
LOW	mcore_adapter/src/mcore_adapter/checkpointing.py	41	if not release:	COMMENT
LOW	mcore_adapter/src/mcore_adapter/platforms/platform.py	41	# available dispatch keys:	COMMENT
LOW	…src/mcore_adapter/models/qwen3_vl/transformer_block.py	241	# size (mbs) == 1, since in this case, 'hidden_states.transpose()'	COMMENT
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	441		COMMENT
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	461	# maxDets - [1 10 100] M=3 thresholds on max detections per image	COMMENT
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	481	# date - date evaluation was performed	COMMENT
LOW	…l/pipeline/rlvr/rewards/remote_reward_system_worker.py	1	# Copyright 2025 alibaba-inc. and/or its affiliates	COMMENT
LOW	roll/pipeline/agentic/agentic_pipeline.py	321		COMMENT
LOW	roll/platforms/platform.py	41	# use "CPU" as a fallback for platforms not registered in PyTorch	COMMENT
LOW	roll/distributed/strategy/fsdp2_strategy.py	241		COMMENT
LOW	roll/utils/taskgroups.py	1	# borrow from cpython	COMMENT
LOW	roll/utils/taskgroups.py	121	# # If this flag is set we must call uncancel().	COMMENT
LOW	roll/utils/taskgroups.py	281	# "not cancelled" later in __aexit__. Example situation that	COMMENT
LOW	roll/third_party/fsdp2/tiled_mlp.py	1	# Copyright 2024 Bytedance Ltd. and/or its affiliates	COMMENT
LOW	roll/third_party/vllm/fp8.py	21		COMMENT
LOW	roll/third_party/megatron/fused_entropy.py	841	)	COMMENT
LOW	roll/third_party/megatron/fused_entropy.py	861		COMMENT
LOW	tests/pipeline/rlvr_megatron_config.yaml	181	std_threshold: 0	COMMENT
LOW	tests/pipeline/rlvr_megatron_config.yaml	201	# data_args:	COMMENT
LOW	tests/pipeline/agentic_pipeline_config.yaml	21	#track_with: wandb	COMMENT
LOW	…distributed/strategy/log_probs/layer_states_capture.py	61	return	COMMENT
LOW	tests/third_party/vllm/test_vllm_mem_oom.py	61	# snapshot_last = tracemalloc.take_snapshot()	COMMENT
LOW	tests/third_party/vllm/test_vllm_mem_oom.py	81	#	COMMENT
LOW	…2.5-7B-agentic_megatron/agentic_val_webshop_gigpo.yaml	21	# name: ${exp_name}_webshop	COMMENT
LOW	…s/qwen2.5-7B-agentic_megatron/agentic_val_webshop.yaml	21	# name: ${exp_name}_webshop	COMMENT
LOW	…vr_megatron/rlvr_config_8gpus_llm_as_judge_server.yaml	81	reward_pretrain: Qwen/Qwen2.5-7B	COMMENT
LOW	…5-7B-rlvr_megatron/rlvr_config_8gpus_llm_as_judge.yaml	81	reward_pretrain: Qwen/Qwen2.5-7B	COMMENT
LOW	…s/qwen3-next-80BA3B-rlvr_megatron/rlvr_config_amd.yaml	81	reward_pretrain: Qwen/Qwen3-Next-80B-A3B-Instruct	COMMENT
LOW	…mples/qwen3-next-80BA3B-rlvr_megatron/rlvr_config.yaml	81	reward_pretrain: Qwen/Qwen3-Next-80B-A3B-Instruct	COMMENT
LOW	…mples/qwen3-30BA3B-rlvr_megatron/rlvr_config_lora.yaml	241	# # NOTE: llm as judge 也需要gpu, 不能和actor infer共享gpu	COMMENT
LOW	…mples/qwen3-30BA3B-rlvr_megatron/rlvr_config_lora.yaml	261	# strategy_args:	COMMENT
LOW	examples/qwen3-omni/audio_test_80G.yaml	41	use_kl_loss: true	COMMENT
LOW	examples/qwen3-omni/audio_test_80G.yaml	81	# "audios": ["dafd0ef215384e8b45326f43ee868204.wav"],	COMMENT
LOW	…mples/agentic_demo/run_openreward_endless_terminals.sh	1	#!/bin/bash	COMMENT
LOW	examples/qwen3_agentic_gem/start_retrieval_server.sh	1	#!/bin/bash	COMMENT
LOW	examples/qwen2.5-7B-rlvr-offpolicy/rlvr_config.yaml	21	# tracker_kwargs:	COMMENT
LOW	examples/qwen2.5-7B-rlvr-offpolicy/rlvr_config.yaml	201	# crossthinkqa:	COMMENT
LOW	examples/qwen2.5-7B-rlvr-offpolicy/rlvr_config.yaml	221	# infer_batch_size: 4	COMMENT
LOW	examples/qwen2.5-7B-rlvr-offpolicy/rlvr_config.yaml	241	# judge_prompt: Qwen2.5-7B-Instruct-RLVR-prompt	COMMENT
LOW	examples/qwen2.5-7B-rlvr-offpolicy/rlvr_config.yaml	261	# strategy_config: null	COMMENT

Redundant / Tautological Comments22 hits · 33 pts

Severity	File	Line	Snippet	Context
LOW	…src/mcore_adapter/parallel_functions/vocab_parallel.py	85	# Check if we can use fused kernel (only for TP=1)	COMMENT
LOW	…core_adapter/parallel_functions/fused_cross_entropy.py	202	# Set loss to 0 if this index should be ignored, otherwise compute normally	COMMENT
LOW	…core_adapter/parallel_functions/fused_cross_entropy.py	456	# Set dloss to 0 if this index should be ignored	COMMENT
LOW	mcore_adapter/src/mcore_adapter/trainer/trainer.py	719	# Check if continuing training from a checkpoint	COMMENT
LOW	roll/pipeline/rlvr/rewards/detection_reward_worker.py	1013	# Check if boxes overlap	COMMENT
LOW	roll/pipeline/agentic/agentic_pipeline.py	937	# Check if this DP rank's GPUs are NOT in the freed set	COMMENT
LOW	roll/pipeline/agentic/utils.py	489	# Check if this teacher has a routing mask (set by pipeline when needs_teacher_routing)	COMMENT
LOW	roll/pipeline/agentic/tools/mcp_tool.py	374	# Check if the item is a text block and has non-empty text	STRING
LOW⚡	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	704	# Check if clean_dir is empty and remove it if so	COMMENT
LOW	roll/pipeline/agentic/proxy/router/alb_proxy_router.py	951	# Add rules to listener and save rule ids to file	COMMENT
LOW	roll/distributed/scheduler/protocol.py	931	# Check if any data has this key, not just the first one	COMMENT
LOW	roll/distributed/scheduler/rollout_scheduler.py	173	# Check if this episode has been submitted	COMMENT
LOW	roll/distributed/strategy/vllm_strategy.py	170	# Check if beam search is requested	COMMENT
LOW	roll/utils/functionals.py	1117	# Check if this teacher has a routing mask (set by pipeline when needs_teacher_routing)	COMMENT
LOW	roll/utils/train_infer_corrections.py	204	# Check if required fields are present	COMMENT
LOW	roll/utils/train_infer_corrections.py	244	# Check if mask is already sliced (shape [B, T-1]) or full (shape [B, T])	COMMENT
LOW	roll/third_party/vllm/gdn_patcher.py	47	# Check if already patched (has _original_build attribute)	COMMENT
LOW	roll/third_party/vllm/gdn_patcher.py	52	# Check if the fix is already in the source	COMMENT
LOW	…ted/strategy/log_probs/test_fsdp_vlm_log_probs_perf.py	212	# Check if results are numerically equivalent	COMMENT
LOW	…ributed/strategy/log_probs/analyze_layer_divergence.py	40	# Check if it's a layer_states file	COMMENT
LOW	…ributed/strategy/log_probs/analyze_layer_divergence.py	168	# Check if this is a nested structure (layer states) or flat (embeddings)	COMMENT
LOW	…ributed/strategy/log_probs/analyze_layer_divergence.py	518	# Check if embeddings show significant divergence	COMMENT

Verbosity Indicators13 hits · 21 pts

Severity	File	Line	Snippet	Context
LOW	roll/pipeline/agentic/agentic_pipeline.py	268	# Step 0: active_dp_ranks initialized with all ranks {0,1,2,3}, no expand needed	COMMENT
LOW⚡	roll/pipeline/agentic/tools/mcp_tool.py	339	# Step 1: Find the schema for the requested tool.	STRING
LOW⚡	roll/pipeline/agentic/tools/mcp_tool.py	347	# Step 2: Use the jsonschema library to validate the parameters.	STRING
LOW⚡	roll/utils/sequence_packing.py	206	# Step 4: Round up to be divisible by vp_size	COMMENT
LOW⚡	roll/utils/sequence_packing.py	210	# Step 5: Calculate workload for load balancing	COMMENT
LOW⚡	roll/utils/sequence_packing.py	215	# Step 6: Partition samples into micro batches with balanced workload	COMMENT
LOW⚡	roll/utils/sequence_packing.py	222	# Step 7: Sort and reorder for better pipeline scheduling	COMMENT
LOW	roll/utils/sequence_packing.py	179	# Step 2: Calculate initial number of micro batches	COMMENT
LOW	roll/utils/sequence_packing.py	192	# Step 3: Synchronize across DP ranks (all ranks must have same count)	COMMENT
LOW	roll/utils/sequence_packing.py	245	# Step 8: Generate micro batches	COMMENT
LOW	roll/configs/base_config.py	690	# Step 1: Normalize teacher to _reference_configs dict (always Dict[str, WorkerConfig])	COMMENT
LOW	roll/configs/base_config.py	711	# Step 2: Pure OPD mode	COMMENT
LOW	roll/configs/base_config.py	732	# Step 3: Mixed OPD mode	COMMENT

AI Slop Vocabulary8 hits · 16 pts

Severity	File	Line	Snippet	Context
LOW	mcore_adapter/src/mcore_adapter/trainer/trainer.py	879	# if loss is nan or inf simply add the average of previous logged losses	COMMENT
MEDIUM	roll/pipeline/agentic/tools/mcp_tool.py	213	"""Generates a comprehensive prompt using the cached tool metadata."""	STRING
LOW	roll/third_party/vllm/ray_distributed_executor.py	134	# No need to sort, just use the given resource order of the placement group	COMMENT
LOW	…ird_party/vllm/vllm_0_12_0/ray_distributed_executor.py	112	# No need to sort, just use the given resource order of the placement group	COMMENT
MEDIUM	…ted/strategy/log_probs/test_fsdp_vlm_log_probs_perf.py	273	"""Test VLM logprobs with CP2 and comprehensive performance statistics."""	STRING
MEDIUM	tests/third_party/vllm/vllm_generate_test.py	36	# "content": "Your role as an assistant involves thoroughly exploring questions through a systematic long think	COMMENT
MEDIUM⚡	…s/docusaurus-plugin-content-docs/current/ROLL_map.html	24	<script src="https://cdn.jsdelivr.net/npm/d3@7.9.0/dist/d3.min.js"></script><script src="https://cdn.jsdelivr.net/npm/ma	CODE
MEDIUM⚡	…s/docusaurus-plugin-content-docs/current/ROLL_map.html	24	<script src="https://cdn.jsdelivr.net/npm/d3@7.9.0/dist/d3.min.js"></script><script src="https://cdn.jsdelivr.net/npm/ma	CODE

Dead Code1 hit · 2 pts

Severity	File	Line	Snippet	Context
MEDIUM	tests/distributed/executor/test_ray_debugger.py	20		CODE

Example Usage Blocks1 hit · 2 pts

Severity	File	Line	Snippet	Context
LOW	…/openreward_endless_terminals_reinforce_qwen35_2b.yaml	5	# Usage:	COMMENT

Fake / Example Data1 hit · 1 pts

Severity	File	Line	Snippet	Context
LOW	…docs/current/User Guides/Agentic/prefix_aggregation.md	139	[*base_chat_history, msg], # base_chat_history = [dummy_sys, dummy_user]	CODE

Analysis Overview

What These Metrics Mean

Score History

Severity Breakdown

Directory Score Breakdown

Pattern Findings