modelscope/ms-swift

Pattern Findings

1935 matches across 17 categories. Click a row to expand file-level details.

Unused Imports877 hits · 760 pts

Severity	File	Line
LOW	setup.py	7
LOW	tests/llm/test_template.py	49
LOW	tests/models/test_mllm.py	7
LOW	tests/models/test_mllm.py	7
LOW	tests/train/test_resume_from_checkpoint.py	29
LOW	tests/train/test_resume_from_checkpoint.py	29
LOW	tests/train/test_freeze.py	17
LOW	tests/train/test_freeze.py	17
LOW	tests/train/test_freeze.py	34
LOW	tests/train/test_freeze.py	34
LOW	tests/train/test_freeze.py	51
LOW	tests/train/test_freeze.py	51
LOW	tests/train/test_freeze.py	68
LOW	tests/train/test_freeze.py	68
LOW	tests/test_align/test_padding_side.py	50
LOW	tests/hub/test_check_model.py	17
LOW	swift/__init__.py	7
LOW	swift/__init__.py	7
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	8
LOW	swift/__init__.py	11
LOW	swift/__init__.py	11
LOW	swift/__init__.py	12
LOW	swift/__init__.py	12
LOW	swift/__init__.py	13
LOW	swift/__init__.py	13
LOW	swift/__init__.py	13
LOW	swift/__init__.py	13
LOW	swift/__init__.py	13
LOW	swift/__init__.py	13
LOW	swift/__init__.py	13
LOW	swift/__init__.py	13
LOW	swift/__init__.py	13
LOW	swift/__init__.py	13
LOW	swift/__init__.py	15
LOW	swift/__init__.py	15
LOW	swift/__init__.py	16
LOW	swift/__init__.py	16
LOW	swift/__init__.py	16
LOW	swift/__init__.py	16
LOW	swift/__init__.py	16
LOW	swift/__init__.py	17
LOW	swift/__init__.py	17
LOW	swift/__init__.py	17
LOW	swift/__init__.py	18
LOW	swift/__init__.py	18
LOW	swift/__init__.py	19
LOW	swift/__init__.py	19
LOW	swift/__init__.py	20
LOW	swift/__init__.py	20
LOW	swift/__init__.py	20
817 more matches not shown…

Hyper-Verbose Identifiers362 hits · 363 pts

Severity	File	Line	Snippet
LOW	tests/test_utils.py	74	def create_dummy_test_dataset(feat, label, num):
LOW	tests/run.py	77	def gather_test_suites_in_files(test_dir, case_file_list, list_tests):
LOW	tests/run.py	132	def async_run_command_with_popen(cmd, device_id):
LOW	tests/run.py	315	def run_non_parallelizable_test_suites(suites, result_dir):
LOW	tests/utils/test_async_rewards.py	17	def test_start_and_shutdown_event_loop_in_daemon(self):
LOW	tests/utils/test_async_rewards.py	38	def test_run_async_function_in_daemon_loop(self):
LOW	tests/utils/test_async_rewards.py	57	def test_async_orm_base_class(self):
LOW	tests/utils/test_async_rewards.py	80	def test_async_reward_is_detected(self):
LOW	tests/utils/test_async_rewards.py	106	def test_parallel_async_execution(self):
LOW	tests/utils/test_async_rewards.py	153	def test_async_reward_function_batch_performance(self):
LOW	tests/utils/test_rewards.py	38	def test_multiple_steps_with_boxed(self):
LOW	tests/utils/test_rewards.py	62	def test_batch_processing_no_tag(self):
LOW	tests/utils/test_rewards.py	73	def test_answer_tag_with_plain_number(self):
LOW	tests/utils/test_rewards.py	82	def test_answer_tag_with_latex(self):
LOW	tests/utils/test_rewards.py	91	def test_long_text_with_answer_tag(self):
LOW	tests/utils/test_rewards.py	106	def test_answer_tag_with_complex_expression(self):
LOW	tests/utils/test_rewards.py	115	def test_solution_with_answer_tag(self):
LOW	tests/utils/test_rewards.py	124	def test_answer_tag_wrong_answer(self):
LOW	tests/utils/test_rewards.py	133	def test_mixed_batch_with_and_without_tags(self):
LOW	tests/utils/test_rewards.py	170	def test_answer_tag_with_extra_whitespace(self):
LOW	tests/utils/test_rewards.py	179	def test_multiple_answer_tags(self):
LOW	tests/utils/test_rewards.py	188	def test_real_world_example_from_user(self):
LOW	tests/utils/test_rewards.py	212	def test_equivalent_fractions(self):
LOW	tests/utils/test_rewards.py	221	def test_different_forms_same_answer(self):
LOW	tests/utils/test_rewards.py	230	def test_latex_inline_math_delimiters(self):
LOW	tests/utils/test_rewards.py	240	def test_latex_display_math_delimiters(self):
LOW	tests/utils/test_rewards.py	249	def test_mixed_latex_delimiters(self):
LOW	tests/tuners/test_swift_device_map.py	25	def test_swift_multiple_adapters(self):
LOW	tests/tuners/test_swift_base.py	73	def test_swift_adapter_forward(self):
LOW	tests/tuners/test_swift_base.py	93	def test_swift_prompt_forward(self):
LOW	tests/tuners/test_swift_base.py	110	def test_swift_restuner_forward(self):
LOW	tests/tuners/test_swift_base.py	133	def lora_injection_with_dtype(self, dtype=torch.float32):
LOW	tests/tuners/test_swift_base.py	185	def test_swift_lora_injection(self):
LOW	tests/tuners/test_swift_base.py	257	def test_swift_multiple_adapters(self):
LOW	tests/tuners/test_swift_base.py	382	def test_swift_multiple_adapters_switching(self):
LOW	tests/tuners/test_peft.py	120	def test_peft_adalora_injection(self):
LOW	tests/tuners/test_extra_state_dict.py	26	def test_swift_extra_state_dict(self):
LOW	tests/tuners/test_extra_state_dict.py	42	def test_swift_modules_to_save(self):
LOW	tests/tuners/test_swift_restuning.py	95	def test_swift_restuning_diffusers_sd(self):
LOW	tests/tuners/test_scetuning.py	64	def test_scetuning_part_mixin(self):
LOW	tests/train/test_vllm_importance_sampling_basic.py	32	def _compute_sequence_level_ratios(self, is_ratio: torch.Tensor, completion_mask: torch.Tensor) -> torch.Tensor:
LOW	tests/train/test_vllm_importance_sampling_basic.py	49	def _apply_rollout_importance_sampling(self, rollout_log_ratio: torch.Tensor,
LOW	tests/train/test_vllm_importance_sampling_basic.py	99	def _compute_is_correction_metrics(
LOW	tests/train/test_vllm_importance_sampling_basic.py	166	def test_token_truncate_basic(self):
LOW	tests/train/test_vllm_importance_sampling_basic.py	199	def test_sequence_truncate_basic(self):
LOW	tests/train/test_vllm_importance_sampling_basic.py	237	def test_threshold_sensitivity(self):
LOW	tests/train/test_vllm_importance_sampling_basic.py	345	def test_clipped_frac_token_truncate(self):
LOW	tests/train/test_vllm_importance_sampling_basic.py	359	def test_clipped_frac_token_mask(self):
LOW	tests/train/test_vllm_importance_sampling_basic.py	373	def test_clipped_frac_sequence_level(self):
LOW	tests/train/test_vllm_importance_sampling_basic.py	391	def test_kl_divergence_same_policy(self):
LOW	tests/train/test_resume_from_checkpoint.py	28	def test_resume_from_checkpoint():
LOW	tests/train/test_sft.py	162	def test_mllm_streaming_mp_ddp():
LOW	tests/train/test_sft.py	215	def test_resume_from_checkpoint():
LOW	tests/train/test_sft.py	275	def test_predict_with_generate():
LOW	tests/train/test_export_cached_dataset.py	1	def test_export_cached_dataset():
LOW	tests/test_align/test_template/test_vision.py	1063	def _infer_ernie_vl_thinking_hf(model, processor, messages):
LOW	tests/test_align/test_template/test_gene.py	7	def test_deepseek_janus_pro_gene():
LOW	docs/source/BestPractices/MLLM-Registration.md	218	def _get_new_tokens_use_audio_in_video(self, i, *, video_grid_thw, video_second_per_grid, audio_lengths,
LOW	docs/source_en/BestPractices/MLLM-Registration.md	221	def _get_new_tokens_use_audio_in_video(self, i, *, video_grid_thw, video_second_per_grid, audio_lengths,
LOW	swift/metrics/reranker.py	111	def calculate_ndcg_single_query(relevance_scores, ranking):
302 more matches not shown…

Deep Nesting326 hits · 298 pts

Severity	File	Line
LOW	setup.py	24
LOW	setup.py	44
LOW	setup.py	78
LOW	setup.py	96
LOW	tests/test_utils.py	124
LOW	tests/run.py	33
LOW	tests/run.py	77
LOW	tests/run.py	185
LOW	tests/run.py	204
LOW	tests/run.py	341
LOW	tests/run.py	454
LOW	tests/general/test_arch.py	1
LOW	tests/train/test_vllm_importance_sampling_basic.py	49
LOW	swift/ui/base.py	268
LOW	swift/ui/base.py	314
LOW	swift/ui/llm_export/llm_export.py	101
LOW	swift/ui/llm_grpo/llm_grpo.py	216
LOW	swift/ui/llm_grpo/external_rollout.py	145
LOW	swift/ui/llm_grpo/tuner.py	18
LOW	swift/ui/llm_rlhf/llm_rlhf.py	232
LOW	swift/ui/llm_rlhf/tuner.py	18
LOW	swift/ui/llm_sample/llm_sample.py	166
LOW	swift/ui/llm_eval/llm_eval.py	104
LOW	swift/ui/llm_infer/runtime.py	133
LOW	swift/ui/llm_infer/runtime.py	265
LOW	swift/ui/llm_infer/llm_infer.py	124
LOW	swift/ui/llm_infer/llm_infer.py	189
LOW	swift/ui/llm_train/task.py	64
LOW	swift/ui/llm_train/runtime.py	295
LOW	swift/ui/llm_train/runtime.py	374
LOW	swift/ui/llm_train/runtime.py	407
LOW	swift/ui/llm_train/runtime.py	542
LOW	swift/ui/llm_train/runtime.py	600
LOW	swift/ui/llm_train/llm_train.py	242
LOW	swift/ui/llm_train/llm_train.py	325
LOW	swift/ui/llm_train/llm_train.py	551
LOW	swift/ui/llm_train/optimizer.py	121
LOW	swift/ui/llm_train/tuner.py	290
LOW	swift/loss/embedding.py	88
LOW	swift/loss/embedding.py	115
LOW	swift/callbacks/activation_cpu_offload.py	330
LOW	swift/callbacks/activation_cpu_offload.py	597
LOW	swift/dataset/dataset_meta.py	120
LOW	swift/dataset/packing.py	32
LOW	swift/dataset/loader.py	70
LOW	swift/dataset/dataset/mllm.py	567
LOW	swift/dataset/dataset/mllm.py	239
LOW	swift/dataset/dataset/mllm.py	715
LOW	swift/dataset/dataset/mllm.py	791
LOW	swift/dataset/dataset/mllm.py	952
LOW	swift/dataset/dataset/mllm.py	1233
LOW	swift/dataset/dataset/llm.py	764
LOW	swift/dataset/preprocessor/core.py	170
LOW	swift/dataset/preprocessor/core.py	491
LOW	swift/rewards/orm.py	238
LOW	swift/rlhf_trainers/rollout_mixin.py	194
LOW	swift/rlhf_trainers/rollout_mixin.py	334
LOW	swift/rlhf_trainers/rollout_mixin.py	471
LOW	swift/rlhf_trainers/rollout_mixin.py	703
LOW	swift/rlhf_trainers/rollout_mixin.py	757
266 more matches not shown…

Excessive Try-Catch Wrapping171 hits · 192 pts

Severity	File	Line	Snippet
LOW	tests/run.py	349	except Exception:
LOW	tests/model_tag.py	66	except Exception as e:
LOW	tests/model_tag.py	94	except Exception as e:
LOW	tests/model_tag.py	115	except Exception as e:
LOW	tests/model_tag.py	131	except Exception as e:
MEDIUM	tests/model_tag.py	53	def _post_request(self, url, param):
MEDIUM	tests/model_tag.py	72	def batch_commit_result(self):
MEDIUM	tests/model_tag.py	100	def batch_refresh_stage(self):
MEDIUM	tests/model_tag.py	121	def query_model_stage(self):
LOW	tests/general/test_arch.py	18	except Exception:
LOW	tests/deploy/test_dataset.py	12	except Exception:
LOW	tests/train/test_vllm_importance_sampling_basic.py	477	except Exception as e:
LOW	docs/source/BestPractices/GRPO.md	78	except Exception as e:
LOW	docs/source/BestPractices/GRPO-Multi-Modal-Training.md	79	except Exception:
LOW	docs/source/BestPractices/GRPO-Multi-Modal-Training.md	96	except Exception:
LOW	…source/Instruction/GRPO/DeveloperGuide/reward_model.md	118	except Exception as e:
LOW	docs/source_en/BestPractices/GRPO.md	82	except Exception as e:
LOW	…s/source_en/BestPractices/GRPO-Multi-Modal-Training.md	85	except Exception:
LOW	…s/source_en/BestPractices/GRPO-Multi-Modal-Training.md	102	except Exception:
LOW	…rce_en/Instruction/GRPO/DeveloperGuide/reward_model.md	117	except Exception as e:
LOW	swift/ui/base.py	347	except Exception as e:
LOW	swift/ui/base.py	368	except Exception:
LOW	swift/ui/llm_grpo/external_runtime.py	135	except Exception as e:
LOW	swift/ui/llm_eval/eval.py	103	except Exception as e:
LOW	swift/ui/llm_infer/runtime.py	259	except Exception as e:
LOW	swift/ui/llm_train/runtime.py	590	except Exception as e:
LOW	swift/ui/llm_train/llm_train.py	406	except Exception as e:
LOW	swift/ui/llm_train/llm_train.py	530	except Exception as err:
LOW	swift/dataset/utils.py	95	except Exception as e:
LOW	swift/dataset/packing.py	151	except Exception as e:
LOW	swift/dataset/loader.py	116	except Exception as e:
LOW	swift/dataset/dataset/mllm.py	796	except Exception:
LOW	swift/dataset/preprocessor/core.py	192	except Exception as e:
LOW	swift/rewards/prm.py	90	except Exception:
LOW	swift/rewards/prm.py	148	except Exception:
LOW	swift/rewards/orm.py	114	except Exception:
LOW	swift/rewards/orm.py	251	except Exception:
LOW	swift/rewards/orm.py	258	except Exception:
LOW	swift/rewards/orm.py	300	except Exception:
LOW	swift/rewards/orm.py	374	except Exception:
LOW	swift/rewards/orm.py	419	except Exception:
LOW	swift/rewards/rm_plugin.py	224	except Exception as e:
MEDIUM	swift/rlhf_trainers/rollout_mixin.py	1563	def infer_task():
MEDIUM	swift/rlhf_trainers/rollout_mixin.py	1573	def done(future):
LOW	swift/rlhf_trainers/rollout_mixin.py	1567	except Exception as e:
LOW	swift/rlhf_trainers/rollout_mixin.py	1577	except Exception as e:
LOW	swift/rlhf_trainers/rollout_mixin.py	1760	except Exception as e:
LOW	swift/rlhf_trainers/reward_trainer.py	102	except Exception as e:
LOW	swift/rlhf_trainers/gkd_trainer.py	1221	except Exception:
LOW	swift/rlhf_trainers/gkd_trainer.py	1273	except Exception as e:
MEDIUM	swift/rlhf_trainers/gkd_trainer.py	1230	def _fetch_one(idx):
LOW	swift/rlhf_trainers/utils.py	1012	except Exception as e: # noqa: BLE001
LOW	swift/rlhf_trainers/utils.py	1099	except Exception as e:
LOW	swift/rlhf_trainers/vllm_client.py	98	except Exception:
LOW	swift/rlhf_trainers/vllm_client.py	172	except Exception as e:
LOW	swift/rlhf_trainers/vllm_client.py	244	except Exception as e:
LOW	swift/rlhf_trainers/vllm_client.py	292	except Exception as e:
LOW	swift/rlhf_trainers/vllm_client.py	354	except Exception as e:
LOW	swift/rlhf_trainers/vllm_client.py	397	except Exception as e:
LOW	swift/rlhf_trainers/vllm_client.py	421	except Exception as e:
111 more matches not shown…

Cross-File Repetition21 hits · 105 pts

Severity	File	Snippet
HIGH	docs/source/BestPractices/GRPO.md	evaluates completions based on mathematical correctness of the answer args: completions (list[str]): generated outputs t
HIGH	docs/source_en/BestPractices/GRPO.md	evaluates completions based on mathematical correctness of the answer args: completions (list[str]): generated outputs t
HIGH	examples/train/grpo/plugin/plugin.py	evaluates completions based on mathematical correctness of the answer args: completions (list[str]): generated outputs t
HIGH	docs/source/BestPractices/GRPO-Multi-Modal-Training.md	reward function that checks if the completion is correct. args: completions (list[str]): generated outputs solution (lis
HIGH	…s/source_en/BestPractices/GRPO-Multi-Modal-Training.md	reward function that checks if the completion is correct. args: completions (list[str]): generated outputs solution (lis
HIGH	examples/train/grpo/plugin/plugin.py	reward function that checks if the completion is correct. args: completions (list[str]): generated outputs solution (lis
HIGH	docs/source/BestPractices/AMD-support.md	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	docs/source/BestPractices/Qwen3_5-Best-Practice.md	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	docs/source/BestPractices/Qwen3_5-Best-Practice.md	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	docs/source_en/BestPractices/AMD-support.md	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	docs/source_en/BestPractices/Qwen3_5-Best-Practice.md	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	docs/source_en/BestPractices/Qwen3_5-Best-Practice.md	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	examples/models/qwen3_5/mcore_grpo_moe.sh	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	examples/train/grpo/plugin/gsm8k/gsm8k.sh	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	examples/train/grpo/internal/fipo.sh	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	examples/train/grpo/internal/real.sh	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	examples/megatron/grpo/real.sh	you are a helpful math assistant. solve the problem step by step and put your final answer within \\boxed{}.
HIGH	docs/source/Instruction/GRPO/DeveloperGuide/gym_env.md	rolloutinferrequest( messages=[ {'role': 'system', 'content': 'a conversation between user and assistant. the user asks
HIGH	…s/source/Instruction/GRPO/DeveloperGuide/multi_turn.md	rolloutinferrequest( messages=[ {'role': 'system', 'content': 'a conversation between user and assistant. the user asks
HIGH	…s/source_en/Instruction/GRPO/DeveloperGuide/gym_env.md	rolloutinferrequest( messages=[ {'role': 'system', 'content': 'a conversation between user and assistant. the user asks
HIGH	…ource_en/Instruction/GRPO/DeveloperGuide/multi_turn.md	rolloutinferrequest( messages=[ {'role': 'system', 'content': 'a conversation between user and assistant. the user asks

Decorative Section Separators32 hits · 99 pts

Severity	File	Line	Snippet
MEDIUM	swift/rlhf_trainers/utils.py	1588	# ============================================================================
MEDIUM	swift/rlhf_trainers/utils.py	1590	# ============================================================================
MEDIUM	swift/rlhf_trainers/grpo_trainer.py	493	# --------------------------------------------------
MEDIUM	swift/rlhf_trainers/grpo_trainer.py	495	# --------------------------------------------------
MEDIUM	swift/rlhf_trainers/grpo_trainer.py	579	# --------------------------------------------------
MEDIUM	swift/rlhf_trainers/grpo_trainer.py	581	# --------------------------------------------------
MEDIUM	swift/pipelines/infer/rollout.py	308	# ------------------------------------------------------------------
MEDIUM	swift/pipelines/infer/rollout.py	310	# ------------------------------------------------------------------
MEDIUM	swift/pipelines/infer/rollout.py	433	# ── Step 1: receive + rebuild IPC handle (with reuse) ────────
MEDIUM	swift/pipelines/infer/rollout.py	471	# ── Step 2: stream buckets and load_weights per bucket ──────
MEDIUM	swift/model/npu_patch/model.py	18	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	20	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	177	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	179	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	209	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	211	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	293	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	295	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	383	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	385	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	429	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	431	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	483	# ---------------------------------------------------------------------------
MEDIUM	swift/model/npu_patch/model.py	485	# ---------------------------------------------------------------------------
MEDIUM	swift/ray/megatron/rollout/ray_vllm_engine.py	136	# ------------------------------------------------------------------
MEDIUM	swift/ray/megatron/rollout/ray_vllm_engine.py	138	# ------------------------------------------------------------------
MEDIUM	swift/ray/megatron/checkpoint_engine/nccl.py	240	# ── Send / Receive ───────────────────────────────────────────────────
MEDIUM	swift/ray/megatron/checkpoint_engine/hccl.py	168	# ── Core lifecycle ───────────────────────────────────────────────────
MEDIUM	swift/ray/megatron/checkpoint_engine/hccl.py	276	# ── Metadata exchange ────────────────────────────────────────────────
MEDIUM	swift/ray/megatron/checkpoint_engine/hccl.py	310	# ── Send / Receive ───────────────────────────────────────────────────
MEDIUM	swift/megatron/arguments/megatron_args.py	170	# ─────────────────────────── Not Supported Yet ───────────────────────────
MEDIUM	examples/train/rlhf/gkd/teacher_server.sh	10	# ========================================================================

Docstring Block Structure14 hits · 70 pts

Severity	File	Line	Snippet
HIGH	docs/README.md	18	Load data from json/yaml/pickle files. This method provides a unified api for loading data from serialized file
HIGH	swift/dataset/loader.py	239	Load and preprocess datasets. This function provides a unified interface to load datasets from various sources (Hug
HIGH	swift/rewards/rm_plugin.py	123	Extract the reward score from the model's output. Args: model_output (str): The model's ou
HIGH	swift/rlhf_trainers/rollout_mixin.py	335	Split model parameters into batches for synchronized weight transfer. This method divides model parameters into
HIGH	swift/rlhf_trainers/rollout_mixin.py	1709	Attempt to encode each input using the template. If encoding fails, resample from a backup iterator unt
HIGH	swift/template/register.py	81	Get or create a template instance for model input/output formatting. This function retrieves the appropriate templa
HIGH	swift/loss_scale/mapping.py	20	Factory function to create a loss scale object from a string specification. The loss_scale string can be in three f
HIGH	swift/utils/hub_utils.py	27	Download model snapshot safely with DDP context protection. This function attempts to download a model from Hugging
HIGH	swift/utils/torch_utils.py	167	Get the last valid (non-padding) token position indices for each sample. This function correctly handles seque
HIGH	swift/rollout/multi_turn.py	30	Perform asynchronous batched inference for multiple rollout requests. This method serves as the main e
HIGH	swift/rollout/multi_turn.py	183	Execute multi-turn conversation rollout with built-in turn management logic. This implements the default multi-
HIGH	swift/model/chunk_gated_delta_rule.py	256	Args: q (torch.Tensor): queries of shape `[B, T, H, K]`. k (torch.Tensor):
HIGH	swift/model/register.py	539	Load a pretrained model and its processor from a model hub or local path. Args: model_id_or_path: The model
HIGH	swift/megatron/trainers/grpo_trainer.py	1576	Attempt to encode each input using the template. If encoding fails, resample from a backup iterator unt

Verbosity Indicators29 hits · 52 pts

Severity	File	Line	Snippet
LOW	tests/utils/test_rewards.py	193	'### Step 1: Recall the formula\n\n'
LOW	tests/utils/test_rewards.py	195	'### Step 2: Use the given terms\n\n'
LOW	tests/utils/test_rewards.py	200	'### Step 3: Find $a_9$\n\n'
LOW	swift/metrics/reranker.py	61	# Step 1: Find all positive sample indices (query boundaries)
LOW	swift/metrics/reranker.py	67	# Step 2: Split into groups (queries)
LOW	swift/metrics/reranker.py	85	# Step 3: Calculate metrics for each query independently
LOW	swift/metrics/reranker.py	135	# Step 4: Calculate mean metrics across all valid queries
LOW	swift/template/templates/minicpm.py	429	# Step 1: Base encode — produces input_ids with -100 for images
LOW	swift/template/templates/minicpm.py	443	# Step 2: Process images — replace -100 tokens with image placeholders
LOW	swift/template/templates/minicpm.py	457	# Step 3: Process audios — expand audio_start/audio_end pairs with <unk> tokens
LOW	swift/template/templates/minicpm.py	495	# Step 4: Compute image_bound using start/end token boundaries
LOW	swift/template/templates/minicpm.py	520	# Step 5: Compute audio_bounds
LOW	swift/sequence_parallel/ulysses.py	511	# Step 1: Gather from all sequence parallel ranks
LOW	swift/sequence_parallel/ulysses.py	519	# Step 2: Gather all rp chunks
LOW	swift/megatron/trainers/vocab_parallel_utils.py	41	# Step 1: Find global max for numerical stability
LOW	swift/megatron/trainers/vocab_parallel_utils.py	45	# Step 2: Compute exp(logits - max) and sum across all TP ranks
LOW	swift/megatron/trainers/vocab_parallel_utils.py	50	# Step 3: Compute log_softmax
LOW	swift/megatron/trainers/grpo_trainer.py	387	# Step 2: Compute KL from logps if kl_in_reward is enabled
LOW	swift/megatron/trainers/grpo_trainer.py	392	# Step 3: Compute advantages (with KL penalty if kl_in_reward is enabled)
LOW	swift/megatron/trainers/grpo_trainer.py	396	# Step 4: Add advantages to encoded batches
LOW	swift/megatron/trainers/grpo_trainer.py	557	# Step 1: Update or append assistant message
LOW	swift/megatron/trainers/grpo_trainer.py	565	# Step 2: Add token IDs and loss mask
LOW	swift/megatron/trainers/grpo_trainer.py	574	# Step 3: Attach rollout extra info
LOW	swift/megatron/trainers/grpo_trainer.py	578	# Step 4: Store finish reason (used for truncation filters etc.)
LOW	swift/megatron/trainers/grpo_trainer.py	583	# Step 5: Store rollout logprobs for importance sampling correction
LOW	swift/megatron/trainers/grpo_trainer.py	591	# Step 6: Store rollout routed_experts for routing replay
LOW	swift/megatron/trainers/grpo_trainer.py	375	# Step 1: Encode batches and compute logps first (unified flow like GRPOTrainer)
LOW	swift/megatron/trainers/grpo_trainer.py	489	# Step 1: Wake up the engine if it's sleeping (vLLM colocate mode)
LOW	swift/megatron/trainers/grpo_trainer.py	496	# Step 2: Load model weights

Self-Referential Comments18 hits · 48 pts

Severity	File	Line	Snippet
MEDIUM	swift/rewards/rm_plugin.py	3	# This module provides plugins for integrating external reward models,
MEDIUM	swift/rewards/rm_plugin.py	169	# Define a mapping for role capitalization if needed
MEDIUM	swift/rlhf_trainers/utils.py	356	# Create a DeepSpeedPlugin with the processed config
MEDIUM	swift/rlhf_trainers/grpo_trainer.py	2014	# Create a copy to avoid modifying the original table used by other loggers.
MEDIUM	swift/template/base.py	461	# This function is only used to handle scenarios where the model needs
MEDIUM	swift/template/base.py	1587	"""This function is important for multi-modal training, as it registers the post_encode method
MEDIUM	swift/pipelines/train/tuner.py	127	"""This function is only useful on the vera tuner"""
MEDIUM	swift/pipelines/infer/rollout.py	144	# Create a stateless process group to manage communication between training processes and vLLM workers.
MEDIUM	swift/pipelines/eval/utils.py	140	# Create a future to receive the result asynchronously
MEDIUM	swift/utils/hf_config.py	13	"""This class is used to read config from config.json(maybe params.json also)"""
MEDIUM	swift/rollout/multi_turn.py	478	# Create a RolloutOutput for the current round
MEDIUM	swift/rollout/multi_turn.py	543	# Create a mock inputs object to use the template's _swift_prepare_inputs method
MEDIUM	examples/train/grpo/plugin/plugin.py	73	# Define a regex pattern that only allows numbers, operators, parentheses, and whitespace
MEDIUM	examples/train/grpo/plugin/plugin.py	220	# Create the sandbox by hand, currently there's no context manager for this version
MEDIUM	examples/train/grpo/plugin/plugin.py	226	# Create a list of tasks for running scripts concurrently
MEDIUM	examples/train/grpo/plugin/plugin.py	205	# Create a new event loop and set it
MEDIUM	examples/train/think_model/qwen3_demo1.sh	3	# This method is also applicable to the Deepseek-R1 series of models.
MEDIUM	examples/custom/my_qwen2_5_omni/my_register.py	307	"""This function is typically used to solve the zero2/zero3 hanging issue in mixed model training,

Over-Commented Block48 hits · 47 pts

Severity	File	Line	Snippet
LOW	tests/general/test_dataset.py	21	'AI-ModelScope/LongAlpaca-12k#1000'
LOW	tests/general/test_dataset.py	81
LOW	tests/train/test_sft.py	421	# test_llm_hqq()
LOW	tests/test_align/test_vllm_vlm.py	201	# test_ovis2()
LOW	tests/test_align/test_template/test_agent.py	721	print(f'labels: {template.safe_decode(encoded2["labels"])}')
LOW	tests/test_align/test_template/test_agent.py	741	# test_glm4_5()
LOW	tests/test_align/test_template/test_llm.py	741	# test_qwen1_5()
LOW	tests/test_align/test_template/test_llm.py	761	# test_phi4()
LOW	tests/test_align/test_template/test_llm.py	781	# test_glm4_5()
LOW	tests/test_align/test_template/test_video.py	401	# test_qwen2_vl()
LOW	tests/test_align/test_template/test_vision.py	1261
LOW	tests/test_align/test_template/test_vision.py	1281	# test_deepseek_vl()
LOW	tests/test_align/test_template/test_vision.py	1301	# test_internvl2_5_mpo()
LOW	tests/test_align/test_template/test_vision.py	1321	# test_glyph()
LOW	tests/test_align/test_template/test_vision.py	1341	# test_mistral_2506()
LOW	tests/test_align/test_template/test_template.py	161
LOW	tests/eval/test_eval.py	61	test_eval_llm()
LOW	docs/source/conf.py	1	# Configuration file for the Sphinx documentation builder.
LOW	docs/source/conf.py	81
LOW	docs/source_en/conf.py	1	# Configuration file for the Sphinx documentation builder.
LOW	docs/source_en/conf.py	81
LOW	docs/source_en/Megatron-SWIFT/LoRA-Training.md	141	# swift export
LOW	swift/rlhf_trainers/args_mixin.py	441	log_rollout_offpolicy_metrics: bool = False # Log off-policy metrics even when IS correction is disabled
LOW	swift/rlhf_trainers/grpo_trainer.py	181	)
LOW	swift/rlhf_trainers/grpo_trainer.py	2601	# 2d. Log PPL difference (sequence-level perplexity difference)
LOW	swift/pipelines/infer/rollout.py	301	"""
LOW	swift/infer_engine/vllm_engine.py	481	else:
LOW	examples/ascend/activation_cpu_offload/train.sh	41	# {'train_runtime': 79.7064, 'train_samples_per_second': 6.311, 'train_steps_per_second': 0.201, 'train_loss': 1.9164841
LOW	examples/deploy/vllm_dp.sh	1	CUDA_VISIBLE_DEVICES=0,1 swift deploy \
LOW	examples/deploy/vllm_dp.sh	21	# "temperature": 0
LOW	examples/deploy/vllm.sh	1	CUDA_VISIBLE_DEVICES=0 swift deploy \
LOW	examples/models/gpt_oss/internvl3_5_gpt.sh	41	--moe_expert_capacity_factor 2 \
LOW	examples/models/qwen3_next/mtp.sh	41	--sequence_parallel true \
LOW	examples/models/qwen3_5/mcore.sh	61	# CUDA_VISIBLE_DEVICES=0,1,2,3 \
LOW	examples/models/qwen3_5/transformers.sh	41	# IMAGE_MAX_TOKEN_NUM=1024 \
LOW	examples/models/qwen3_5/fp8.sh	61	# swift infer \
LOW	…mples/train/multimodal/lora_llm_full_vit/merge_lora.sh	1	CUDA_VISIBLE_DEVICES=0 \
LOW	…s/train/sequence_parallel/sequence_parallel_qwen3_5.sh	21	--logging_steps 1 \
LOW	examples/train/rlhf/opsd/opsd.sh	1	# OPSD Training Script
LOW	examples/train/grpo/plugin/run_external_scheduler.sh	1	# This script require main branch ms-swift
LOW	examples/train/grpo/plugin/deepeyes/deepeyes.sh	1	# 8 * 80G
LOW	examples/train/grpo/internal/full_lmdeploy.sh	1	# The LMDeploy backend in GRPO has been deprecated in Swift 3.5.
LOW	examples/train/grpo/external/vllm_multi_turn.sh	1	# Exp: https://github.com/modelscope/ms-swift/pull/5307#issuecomment-3219803922
LOW	examples/megatron/mcore_bridge/full/dense.sh	41	# VIDEO_MAX_TOKEN_NUM=128 \
LOW	examples/megatron/mcore_bridge/lora/seq_cls.sh	41	--no_save_rng true \
LOW	examples/megatron/multimodal/lora_llm_vit_full/sft.sh	61	# IMAGE_MAX_TOKEN_NUM=1024 \
LOW	examples/megatron/grpo/dense_server.sh	1	# MAX_PIXELS=602112 \
LOW	examples/megatron/lora/mtp.sh	41	--dataloader_num_workers 8 \

Redundant / Tautological Comments21 hits · 34 pts

Severity	File	Line	Snippet
LOW	swift/callbacks/activation_cpu_offload.py	603	# Check if model is wrapped with FSDP
LOW	swift/callbacks/activation_cpu_offload.py	607	# Check if fsdp_config is a dictionary and has activation_cpu_offload enabled
LOW	swift/callbacks/lisa.py	43	# Check if it's time to switch active layers, including at step 0
LOW	swift/rlhf_trainers/rollout_mixin.py	985	# Check if the number of logprobs matches the number of loss_mask=1 tokens
LOW	swift/rlhf_trainers/utils.py	932	# Check if already patched (idempotent).
LOW	swift/rlhf_trainers/utils.py	964	# Check if inner_model has layers attribute
LOW	swift/rlhf_trainers/grpo_trainer.py	2323	# Set max_length to None to disable truncation, as the input length has already been truncated earlier.
LOW	swift/pipelines/eval/utils.py	168	# Check if we've reached the desired batch size
LOW	swift/trainers/mixin.py	406	# Check if we should delete older checkpoint(s)
LOW	swift/trainers/reranker_trainer.py	18	# Check if we have a custom loss function
LOW	swift/rollout/multi_turn.py	289	# Check if the number of logprobs matches the number of loss_mask=1 tokens
LOW	swift/infer_engine/utils.py	539	# Check if we have a cached last_output from the previous iteration.
LOW	swift/infer_engine/utils.py	606	# Check if need to run the usual non-async path
LOW	swift/hub/hub.py	351	# Write the file if it has changed
LOW	swift/arguments/rlhf_args.py	559	# Check if teacher_deepspeed is a predefined name
LOW	swift/megatron/utils/megatron_lm_utils.py	520	# Set bucket_size to infinity if overlap_grad_reduce is False.
LOW	swift/megatron/trainers/grpo_trainer.py	1103	# Check if this is the PP last stage (only last stage has labels and computes loss)
LOW	swift/agent_template/minimax_m2.py	72	# Check if using react format
LOW	examples/train/grpo/plugin/plugin.py	69	# Check if all numbers are used exactly once
LOW	examples/train/grpo/plugin/plugin.py	81	# Check if the equation is correct and matches the ground truth
LOW	examples/train/grpo/plugin/plugin.py	57	# Check if the format is correct

Slop Phrases5 hits · 13 pts

Severity	File	Line	Snippet
LOW	swift/model/models/qwen.py	1687	loss.device) # make sure to reside in the same device
MEDIUM	examples/train/multi-gpu/fsdp2_lora/train.sh	2	# NOTE: for swift>=3.12, you can use --fsdp fsdp2 instead of accelerate launch
MEDIUM	examples/train/multi-gpu/fsdp_qlora/train.sh	2	# NOTE: for swift>=3.12, you can use --fsdp fsdp2 instead of accelerate launch
MEDIUM	examples/megatron/fp8/lora.sh	3	# However, you can use BF16 weights to perform Merge-LoRA.
MEDIUM	examples/megatron/fp8/lora.sh	64	# Alternatively, you can use BF16 base model + BF16 LoRA for inference

Cross-Language Confusion3 hits · 12 pts

Severity	File	Line	Snippet
HIGH	swift/rlhf_trainers/grpo_trainer.py	2535	- kl: Direct KL divergence estimator KL(π_rollout \|\| π_training)
HIGH	swift/hub/hub.py	357	repo.push(commit_message)
HIGH	swift/megatron/trainers/vocab_parallel_utils.py	108	KL(target \|\| input) = sum(target_prob * (target_log_prob - input_log_prob))

Magic Placeholder Names2 hits · 10 pts

Severity	File	Line	Snippet
HIGH	docs/source/Instruction/Sample.md	80	OPENAI_API_KEY="your_api_key" \
HIGH	docs/source_en/Instruction/Sample.md	84	OPENAI_API_KEY="your_api_key" \

AI Slop Vocabulary4 hits · 9 pts

Severity	File	Line	Snippet
LOW	tests/train/test_vllm_importance_sampling_basic.py	20	# In testing, just return the tensor as-is
MEDIUM	swift/callbacks/perf_log.py	58	# TODO Collect comprehensive TFLOPS data. Then provide a fallback strategy based on lookup tables.
MEDIUM	swift/template/templates/minicpm.py	496	# This is more robust than finding consecutive <unk> tokens, especially
LOW	examples/megatron/export/lora.sh	3	# simply set `--merge_lora true`

Synthetic Comment Markers1 hit · 5 pts

Severity	File	Line	Snippet
HIGH	…s/source_en/BestPractices/GRPO-Multi-Modal-Training.md	278	This task is based on the experiments in [open-r1-multimodal](https://github.com/EvolvingLMMs-Lab/open-r1-multimodal.git

Overly Generic Function Names1 hit · 1 pts

Severity	File	Line	Snippet
LOW	swift/callbacks/activation_cpu_offload.py	490	def my_function(*inputs):

Score History

Severity Breakdown

Pattern Findings